340 33 20MB
English Pages XXII, 488 [491] Year 2021
Lecture Notes in Bioengineering
Albert A. Rizvanov Bikesh Kumar Singh Padma Ganasala Editors
Advances in Biomedical Engineering and Technology Select Proceedings of ICBEST 2018
Lecture Notes in Bioengineering Advisory Editors Nigel H. Lovell, Graduate School of Biomedical Engineering, University of New South Wales, Kensington, NSW, Australia Luca Oneto, DIBRIS, Università di Genova, Genova, Italy Stefano Piotto, Department of Pharmacy, University of Salerno, Fisciano, Italy Federico Rossi, Department of Earth, University of Salerno, Fisciano, Siena, Italy Alexei V. Samsonovich, Krasnow Institute for Advanced Study, George Mason University, Fairfax, VA, USA Fabio Babiloni, Department of Molecular Medicine, University of Rome Sapienza, Rome, Italy Adam Liwo, Faculty of Chemistry, University of Gdansk, Gdansk, Poland Ratko Magjarevic, Faculty of Electrical Engineering and Computing, University of Zagreb, Zagreb, Croatia
Lecture Notes in Bioengineering (LNBE) publishes the latest developments in bioengineering. It covers a wide range of topics, including (but not limited to): • • • • • • • • • • •
Bio-inspired Technology & Biomimetics Biosensors Bionanomaterials Biomedical Instrumentation Biological Signal Processing Medical Robotics and Assistive Technology Computational Medicine, Computational Pharmacology and Computational Biology Personalized Medicine Data Analysis in Bioengineering Neuroengineering Bioengineering Ethics
Original research reported in proceedings and edited books are at the core of LNBE. Monographs presenting cutting-edge findings, new perspectives on classical fields or reviewing the state-of-the art in a certain subfield of bioengineering may exceptionally be considered for publication. Alternatively, they may be redirected to more specific book series. The series’ target audience includes advanced level students, researchers, and industry professionals working at the forefront of their fields. Indexed by SCOPUS and Springerlink. The books of the series are submitted for indexing to Web of Science.
More information about this series at http://www.springer.com/series/11564
Albert A. Rizvanov Bikesh Kumar Singh Padma Ganasala •
•
Editors
Advances in Biomedical Engineering and Technology Select Proceedings of ICBEST 2018
123
Editors Albert A. Rizvanov Kazan Federal University Kazan, Tatarstan Republic, Russia
Bikesh Kumar Singh Department of Biomedical Engineering National Institute of Technology Raipur Raipur, India
Padma Ganasala Gayatri Vidya Parishad College of Engineering Visakhapatnam, Andhra Pradesh, India
ISSN 2195-271X ISSN 2195-2728 (electronic) Lecture Notes in Bioengineering ISBN 978-981-15-6328-7 ISBN 978-981-15-6329-4 (eBook) https://doi.org/10.1007/978-981-15-6329-4 © Springer Nature Singapore Pte Ltd. 2021 This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed. The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use. The publisher, the authors and the editors are safe to assume that the advice and information in this book are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or the editors give a warranty, express or implied, with respect to the material contained herein or for any errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. This Springer imprint is published by the registered company Springer Nature Singapore Pte Ltd. The registered company address is: 152 Beach Road, #21-01/04 Gateway East, Singapore 189721, Singapore
Organising Committee of ICBEST-2018
Chief Patron Prof. A. M. Rawani, Director, NIT Raipur
Patron Prof. S. Sanyal, TEQIP III Coordinator, NIT Raipur
Internal Adivsory Committee Dr. Dr. Dr. Dr. Dr. Dr.
Shrish Verma, Dean Academics, NIT Raipur A. P. Rajimwale, Dean Faculty Welfare, NIT Raipur P. Diwan, Dean, Student Welfare, NIT Raipur (Mrs.) Shubhrata Gupta, Dean, R & C, NIT Raipur G. D. Ramtekkar, Dean, P & D, NIT Raipur P. Y. Dhekne, Registrar, NIT Raipur
Chairman Dr. Bikesh Kumar Singh, Department of Biomedical Engineering, NIT Raipur
v
vi
Organising Committee of ICBEST-2018
Conference Secretaries Dr. Neelamshobha Nirala, Department of Biomedical Engineering, NIT Raipur Dr. Arindam Bit, Department of Biomedical Engineering, NIT Raipur Dr. Saurabh Gupta, Department of Biomedical Engineering, NIT Raipur
Contents
Extraction and Phytochemical Analysis of Coccinia indica Fruit Using UV-VIS and FTIR Spectroscopy . . . . . . . . . . . . . . . . . . . . . . . . . . . . Alok Sharma, Bidyut Mazumdar, and Amit Keshav 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Material and Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Result and Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Numerical Simulation Method to Predict Air Flow and Contaminant Control in a Multiple Bed Intensive Care Unit of Hospital . . . . . . . . Arvind Kumar Sahu, Shobha Lata Sinha, and Tikendra Nath Verma 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Governing Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Problem Statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 Result and Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . A Simple Robust Image Processing Algorithm for Analysis of Static Foot Pressure Intensity Image to Detect Foot Risk Areas in Diabetic Patients . . . . . . . . . . . . . . . . . . . . . . . . . . Hari S. Nair, Navya Thomas, and R. Periyasamy 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1 Foot Image Data Acquisition . . . . . . . . . . . . . . . . . . . 2.2 Image Processing Algorithms . . . . . . . . . . . . . . . . . . 3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
..
1
. . . . .
. . . . .
1 2 3 6 6
..
9
. . . . . .
9 11 12 12 17 18
.........
21
. . . . . . .
22 23 23 24 24 25 28
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . .
. . . . . . .
. . . . . . .
vii
viii
Contents
Heavy Metal Ions Detection by Carbon Paste Electrode as an Electrochemical Sensor . . . . . . . . . . . . . . . . . . . . . Arti Mourya, Bidyut Mazumdar, and Sudip K. Sinha 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Material and Methods . . . . . . . . . . . . . . . . . . . . . . . . . 2.1 Reagents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 Preparation of Electrode as a Sensor . . . . . . . . . . 3 Result and Discussion . . . . . . . . . . . . . . . . . . . . . . . . . 3.1 Cyclic Voltammetry . . . . . . . . . . . . . . . . . . . . . . 3.2 Effect of Operating Parameters . . . . . . . . . . . . . . 3.3 Electrochemical Impedance Spectroscopy (EIS) . . 4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Median Filtering Detection Using Markov in Digital Images . . . . . . . . . . . . . . . . . . . Saurabh Agarwal and Satish Chand 1 Introduction . . . . . . . . . . . . . . . . . . . . . 2 Feature Construction . . . . . . . . . . . . . . . 3 Experimental Setup and Results . . . . . . . 3.1 Datasets . . . . . . . . . . . . . . . . . . . . 3.2 Classifier . . . . . . . . . . . . . . . . . . . 3.3 Experimental Results . . . . . . . . . . 4 Conclusion . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . .
............
29
. . . . . . . . . .
. . . . . . . . . .
29 30 30 30 30 30 32 33 33 34
.......................
35
. . . . . . . .
. . . . . . . .
35 37 38 38 39 39 39 42
..
43
. . . . . .
. . . . . .
44 45 45 48 51 51
..........
53
. . . . . .
54 56 56 58 58 61
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
Process
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
Differential of EMG Activity of Selected Calf Muscle During DLHR Exercise in Relation to Performance Level . . . . . . . . . . . . . . . . . . . . . Monika, L. M. Saini, and Saravjeet Singh 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Material and Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1 Data Acquisition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Result and Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Advanced Encryption Standard Algorithm in Multimodal Biometric Image . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Sharmila S. More, Bhawna Narain, and B. T. Jadhav 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Multimodal Biometrics . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1 Role of Multimodal Biometrics in AES . . . . . . . . . . 3 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1 Proposed Architecture . . . . . . . . . . . . . . . . . . . . . . . 3.2 Module Implementation . . . . . . . . . . . . . . . . . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . . . .
. . . . . .
. . . . . .
Contents
ix
4 Results and Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . In-Silico Construction of Hybrid ORF Protein to Enhance Algal Oil Content for Biofuel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Mohit Nigam, Ruchi Yadav, and Garima Awasthi 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Material and Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Result and Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1 Identification of Algal Gene and Its Functional Protein . . . . . . . 3.2 Hybrid ORF Construction Using Selected Conserved Regions of Superfamilies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3 Hybrid ORF Clone Designed Using Vector NTI Tool Kit . . . . . 3.4 Multi-template Homology Modeling of Hybrid ORF Protein Using Schrödinger Software . . . . . . . . . . . . . . . . . . . . . . . . . . 3.5 Protein Structure Prediction of the Constructed Hybrid ORF by Phyre2 Server . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ANFIS Detects the Changes in Stressful Patterns of Sleep Prabhat Kumar Upadhyay and Chetna Nagpal 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Materials and Methods . . . . . . . . . . . . . . . . . . . . . . . . . 2.1 Features . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 Neuro-fuzzy System . . . . . . . . . . . . . . . . . . . . . . . 2.3 Rules for Manual Scoring . . . . . . . . . . . . . . . . . . . 2.4 Observations on Fuzziness in Stress . . . . . . . . . . . . 3 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1 Data Recording . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2 Feature Selection . . . . . . . . . . . . . . . . . . . . . . . . . 3.3 Sleep Stage Classification . . . . . . . . . . . . . . . . . . . 3.4 Stress Level Classification . . . . . . . . . . . . . . . . . . . 4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
63 65 65
..
67
. . . .
. . . .
68 69 69 69
.. ..
69 70
..
81
.. .. ..
82 87 87
EEG . . . . . .
91
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
Recent Advances in Deep Learning Techniques and Its Applications: An Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Abhishek Hazra, Prakash Choudhary, and M. Sheetal Singh 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Overview of Deep Learning Techniques . . . . . . . . . . . . . . . . . . . . . . 2.1 Convolutional Neural Network . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
. 91 . 93 . 93 . 93 . 94 . 94 . 95 . 95 . 96 . 96 . 98 . 99 . 99 . 100 . 101
. . 103 . . 103 . . 105 . . 105
x
2.2 Recurrent Neural Network . . . . . . . . . . . . . . . . . . . 2.3 Long Short-Term Memory . . . . . . . . . . . . . . . . . . 3 Applications of Deep Learning Techniques . . . . . . . . . . 3.1 Action and Gesture Recognition . . . . . . . . . . . . . . 3.2 Deep Learning for Big Data . . . . . . . . . . . . . . . . . 3.3 Deep Learning for Sentiment Analysis . . . . . . . . . . 3.4 Deep Learning for Medical Image Analysis . . . . . . 3.5 Deep Learning for Text Detection and Recognition 3.6 Deep Learning for Image Classification . . . . . . . . . 3.7 Deep Learning for Object Detection . . . . . . . . . . . . 3.8 Deep Learning for Object Tracking . . . . . . . . . . . . 4 Software and Implementation Tools . . . . . . . . . . . . . . . . 5 Discussion Overview . . . . . . . . . . . . . . . . . . . . . . . . . . 6 Conclusion and Future Work . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Contents
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
Validation of a New Method of Pediatric Refraction: Large Aperture Lens Rack . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Anupam Sahu, Samrat Chatterjee, Deepshikha Agrawal, and Pradeep Chand Dubey 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1 Comparison Between Retinoscopy Values . . . . . . . . . . . . . . . . 3.2 Correlation Between Retinoscopy Values . . . . . . . . . . . . . . . . . 4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.1 Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Comparative Evaluation of in Vitro Antioxidant, Amylase Inhibition and Cytotoxic Activity of Cur-Pip Dual Drug Loaded Nanoparticles . Trilochan Satapathy, Prasanna Kumar Panda, and Gitanjali Mishra 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Materials and Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1 Materials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3 Preparation of Nanoparticles . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4 Amylase Inhibition Assay . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.5 Analysis of Acarbose as Standard Inhibitor . . . . . . . . . . . . . . . 2.6 Determination of Total Antioxidant Capacity . . . . . . . . . . . . . . 2.7 Bacterial Strain-Based Cytotoxicity Screening: . . . . . . . . . . . . . 3 Results and Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1 Amylase Inhibition by Different Nanoformulation . . . . . . . . . . .
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
106 106 107 107 107 108 109 109 111 111 112 112 112 114 115
. . 123
. . . . . . . . .
. . . . . . . . .
123 125 126 126 127 127 128 128 128
. . 129 . . . . . . . . . . .
. . . . . . . . . . .
130 131 131 131 131 132 132 133 134 134 134
Contents
xi
4 Conflict of Interest . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138 Improved ERP Classification Algorithm for Brain–Computer Interface of ALS Patient . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Vyom Raj, Shreya Sharma, Mridu Sahu, and Samrudhi Mohdiwale 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Literature Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Dataset Description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.1 Data Acquisition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2 Feature Extraction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.3 Feature Selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.4 Classification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . .
Size Reduction in Multiband Planar Antenna for Wireless Applications Using Current Distribution Technique . . . . . . Pravin Tajane and P. L. Zade 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2 Proposed Technique . . . . . . . . . . . . . . . . . . . . . . . . 2 Antenna Design and Simulation Approach . . . . . . . . . . . . 3 Simulation Result . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 Experimental Results . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . .
Classification of Hepatic Disease Using Machine Learning Algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Lokesh Singh, Rekh Ram Janghel, and Satya Prakash Sahu 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Related Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Classification Experiments . . . . . . . . . . . . . . . . . . . . . . . . 3.1 System Architecture . . . . . . . . . . . . . . . . . . . . . . . . 3.2 Normalization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3 Classification Algorithms . . . . . . . . . . . . . . . . . . . . 3.4 Dataset Used . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . 141 . . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
141 142 143 143 144 144 145 146 146 147 148
. . . . . . . . . . 151 . . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
152 152 153 153 156 157 157 160
. . . . . . . . . . 161 . . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
162 163 164 164 164 164 167 169 172 173
xii
Contents
Anti-hyperlipidemic and Antioxidant Activities of a Combination of Terminalia Arjuna and Commiphora Mukul on Experimental Animals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Jhakeshwar Prasad, Ashish Kumar Netam, Trilochan Satapathy, S. Prakash Rao, and Parag Jain 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Materials and Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1 Drug and Chemical Reagents . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 Biochemical Estimation of Antioxidants . . . . . . . . . . . . . . . . . . 3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1 The Effect of Terminalia Arjuna Along with Commiphora Mukul on Behavioral Changes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2 The Effect of Terminalia Arjuna Along with Commiphora Mukul on Hematological Changes . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3 Biochemical Parameters Studies . . . . . . . . . . . . . . . . . . . . . . . . 4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Epileptic Seizure Detection Using Deep Recurrent Neural Networks in EEG Signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Archana Verma and Rekh Ram Janghel 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1 Dataset . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 Discrete Wavelet Transform . . . . . . . . . . . . . . . . . . . . . . . . . 2.3 Recurrent Neural Networks (RNNs) . . . . . . . . . . . . . . . . . . . . 2.4 Gated Recurrent Unit (GRU) . . . . . . . . . . . . . . . . . . . . . . . . . 3 Result and Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1 Result . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 Comparison with Other Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 Conclusion and Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . .
Detection of Disease from Leaf of Vegetables and Fruits Learning Technique . . . . . . . . . . . . . . . . . . . . . . . . . . . . Avisha Jaiswal, Saurabh Pathak, Yogesh Kumar Rathore, and Rekh Ram Janghel 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.1 Deep Learning . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Literature Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1 Data Acquisition . . . . . . . . . . . . . . . . . . . . . . . . . 3.2 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . .
. . 175
. . . . .
. . . . .
175 176 176 178 179
. . 179 . . . . .
. . . . .
179 180 181 187 187
. . . 189 . . . . . . . . . . .
. . . . . . . . . . .
189 191 191 191 192 193 196 196 196 196 197
Using Deep . . . . . . . . . . . . 199
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
200 200 201 202 202 202
Contents
xiii
4 Result . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204 5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205 Evaluation for Toxicity and Improved Therapeutic Effectiveness of Natural Polymer Co-administered Along with Venocin in Acetic Acid-Induced Colitis Using Rat Model . . . . . . . . . . . . . Ashish Kumar Netam, Jhakeshwar Prasad, Trilochan Satapathy, and Parag Jain 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Materials and Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1 Drug and Chemical Reagents . . . . . . . . . . . . . . . . . . . . . . 2.2 Experimental Animals . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3 Experimental Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4 Acute Toxicity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.5 Induction of Colitis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.6 Hematological Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.7 Antioxidant Activity Lipid Peroxidase/Malonaldehyde (LPO/MDA) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.8 Measurement of TNF-a . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.9 Histopathological Evaluation . . . . . . . . . . . . . . . . . . . . . . . 3 Result . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1 Behavioral Changes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2 Body Weight Loss . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3 Hematological Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.4 MDA Activity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.5 TNF-a Activity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.6 Histopathology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Finite Element Analysis of Traumatic Brain Injury Due to Blunt Impact of Different Durations . . . . . . . . . . . . . . . . . . . . . . . . . . . Tanu Khanuja and Harikrishnan Narayanan Unni 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Material and Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1 MRI Predicated 3-D Human Head Model and Mesh Generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 Material Properties and Model Validation . . . . . . . . . . . . . 3 Results and Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . 207
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
208 209 209 209 209 210 210 210
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
211 211 211 212 212 212 212 212 216 216 216 218 219
. . . . . 221 . . . . . 221 . . . . . 222 . . . . .
. . . . .
. . . . .
. . . . .
. . . . .
222 224 226 228 228
xiv
Contents
Data Dissemination Using Social-Based Attributes in Delay-Tolerant Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Sanjay Kumar, Prasoon Shukla, and Sudhakar Pandey 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Community . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Betweenness Centrality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 Degree Centrality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 Similarity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 Proposed Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 Performance Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.1 Buffer Utilization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.2 Delivery Ratio . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Hiding Patient Information in Medical Images: A Robust Watermarking Algorithm for Healthcare System . . . . . . . Ritu Agrawal, Manisha Sharma, and Bikesh Kumar Singh 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.1 Contribution and Outline of the Paper . . . . . . . . . . 2 Materials and Methods . . . . . . . . . . . . . . . . . . . . . . . . . 2.1 Brain Image Dataset . . . . . . . . . . . . . . . . . . . . . . . 2.2 Proposed Watermarking Scheme [Embedding and Extraction] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Results and Discussion . . . . . . . . . . . . . . . . . . . . . . . . . 3.1 Performance Measures . . . . . . . . . . . . . . . . . . . . . 3.2 Visual Quality Evaluation . . . . . . . . . . . . . . . . . . . 3.3 Robustness Analysis of DICOM Images . . . . . . . . 4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . 231 . . . . . . . . . . .
. . . . . . . . . . .
231 233 234 235 236 238 239 239 240 242 242
. . . . . . . . . . . 245 . . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
246 249 249 251
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
251 253 253 254 255 260 261
Segmented Lung Boundary Correction in Chest Radiograph Using Context-Aware Adaptive Scan Algorithm . . . . . . . . . . . Tej Bahadur Chandra, Kesari Verma, Deepak Jain, and Satyabhuwan Singh Netam 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Proposed Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Materials and Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1 Dataset . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2 Lung Segmentation . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3 Evaluation Measures . . . . . . . . . . . . . . . . . . . . . . . . . . 4 Experimental Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 Results and Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . 263
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
264 266 268 268 268 269 270 271
Contents
xv
6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 274 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 274 Effect of Temperature and Titania Doping on Structure of Hydroxyapatite . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Yash Chopra, Rajesh Kumar, and Howa Begam 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Material and Method . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1 Materials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Results and Discussion . . . . . . . . . . . . . . . . . . . . . . . . 4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . 277 . . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
Preparation and Characterization of Cellulose Nano Crystal/PVA/ Chitosan Composite Film for Wound Healing Application . . . . . . . Shubham Sen, Rashmi Agrawal, and Howa Begam 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Materials and Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1 Materials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 Preparation of Film . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3 Characterization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4 Statistical Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Results and Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Automated CAD System for Skin Lesion Diagnosis: A Review Lokesh Singh, Rekh Ram Janghel, and Satya Prakash Sahu 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Skin Cancer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1 Melanoma Skin Cancer . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 Non-Melanoma Skin Cancer . . . . . . . . . . . . . . . . . . . . . 3 Image Acquisition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 CAD System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.1 Preprocessing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2 Lesion Segmentation . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.3 Feature Extraction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.4 Feature Analysis and Selection . . . . . . . . . . . . . . . . . . . 4.5 Classification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 Performance of Evaluation Measures . . . . . . . . . . . . . . . . . . . 6 Conclusion and Future Trends . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
277 278 278 279 283 283
. . . . 285 . . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
285 286 286 287 287 289 289 292 292
. . . . . . . 295 . . . . . . . . . . . . . .
. . . . . . . . . . . . . .
. . . . . . . . . . . . . .
. . . . . . . . . . . . . .
. . . . . . . . . . . . . .
. . . . . . . . . . . . . .
. . . . . . . . . . . . . .
295 298 299 300 301 302 302 304 306 307 309 310 316 317
xvi
Contents
Medical Diagnosis of Coronary Artery Disease Using Fuzzy Rule-Based Classification Approach . . . . . . . . . . . . . . . . . . . Namrata Singh and Pradeep Singh 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Fuzzy Rule-Based Methodology . . . . . . . . . . . . . . . . . . . . 2.1 Variables Selection . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 Fuzzification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3 Knowledge Base (IF–THEN Rules Formulation) . . . . 3 Results and Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1 Patients . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2 Comparison with the Prior Work . . . . . . . . . . . . . . . . 4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . .
Segmentation of Lungs in Thoracic CTs Using K-means and Morphological Operations . . . . . . . . . . . . . . . . . . . . Satya Prakash Sahu, Rahul Kumar, Narendra D. Londhe, and Shrish Verma 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Related Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Material and Methods . . . . . . . . . . . . . . . . . . . . . . . . . 3.1 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2 Data Preprocessing . . . . . . . . . . . . . . . . . . . . . . . 3.3 Segmentation of Lungs . . . . . . . . . . . . . . . . . . . . 3.3.1 K-means Algorithm . . . . . . . . . . . . . . . . . 3.3.2 Thresholding . . . . . . . . . . . . . . . . . . . . . . 4 Experimental Results and Analysis . . . . . . . . . . . . . . . 4.1 Evaluation Measures . . . . . . . . . . . . . . . . . . . . . . 4.2 Result Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . 321 . . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
321 323 323 323 323 324 324 324 325 329
Clustering . . . . . . . . . . . . 331
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
Univariate Feature Selection Techniques for Classification of Epileptic EEG Signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Moushmi Kar and Laxmikant Dewangan 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.1 Related Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Material and Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1 Dataset . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 Feature Extractions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3 Feature Selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4 Statistical Measures and Classification of EEG Signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Result and Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
331 332 333 333 333 334 334 336 337 337 338 339 341
. . 345 . . . . . .
. . . . . .
346 347 349 350 350 350
. . 353 . . 355
Contents
4 Conclusion and Future Perspectives . 4.1 Conclusion . . . . . . . . . . . . . . . 4.2 Future Perspectives . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . .
xvii
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
Feature Selection for Classification of Breast Cancer in Histopathology Images: A Comparative Investigation Using Wavelet-Based Color Features . . . . . . . . . . . . . . . . . . . Kushangi Atrey, Bikesh Kumar Singh, and Narendra K. Bodhey 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2 Related Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.3 Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.4 Organization of Paper . . . . . . . . . . . . . . . . . . . . . . . . . 2 Materials and Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 Feature Extraction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3 Feature Selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4 Classification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Results and Discussions . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 Conclusion and Future Scope . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
362 362 363 363
. . . . . . . . 367 . . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
Numerical Study on Particle Deposition in Healthy Human Airways and Airways with Glomus Tumor . . . . . . . . . . . . . . . . . . . . . . . . . . . Digamber Singh, Anuj Jain, and Akshoy Ranjan Paul 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.1 Human Upper Respiratory Tract . . . . . . . . . . . . . . . . . . . . . . . 2 Numerical Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1 Governing Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 Numerical Grid Generation and Solution Schemes . . . . . . . . . . 2.3 Discrete-Phase Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4 Particle Deposition Model Validation . . . . . . . . . . . . . . . . . . . . 3 Results and Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Virtual Reality Therapy in Prolonging Attention Spans for ADHD S. Sushmitha, B. Tanushree Devi, V. Mahesh, B. Geethanjali, K. Arun Kumar, and P. G. Pavithran 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Materials and Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1 Participant Selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 Selection of Tasks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
368 368 368 370 370 371 371 371 371 373 375 376 376
. . 379 . . . . . . . . . .
. . . . . . . . . .
379 381 382 382 383 384 385 386 389 389
. . . . 391
. . . .
. . . .
. . . .
. . . .
391 392 392 393
xviii
2.3 Experimental Protocol . . . . . . 2.4 Signal Processing . . . . . . . . . 3 Results and Discussion . . . . . . . . . 3.1 Visualization of ADHD Brain 3.2 Theta/Alpha Ratio . . . . . . . . . 4 Conclusion . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . .
Contents
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
Multispectral NIRS System Design to Analyze Hemoglobin Concentration on Plantar Foot Surface . . . . . . . . . . . . . . . . . . . Resham Raj Shivwanshi, N. P. Guhan Seshadri, and R. Periyasamy 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Instrumentation and Experiments . . . . . . . . . . . . . . . . . . . . . . . 3 Experiment Protocol . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 Result . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
394 396 396 396 397 399 399
. . . . . . 401 . . . . . . .
Comparative Investigation of Different Classification Techniques for Epilepsy Detection Using EEG Signals . . . . . . . . . . . . . . . . . . Sunandan Mandal, Manvendra Thakur, Kavita Thakur, and Bikesh Kumar Singh 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Materials and Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1 Data Description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 Feature Extraction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3 Feature Selection and Classification Technique . . . . . . . . . . 2.4 Proposed Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.5 Performance Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Results and Discussions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
402 403 405 407 408 411 412
. . . . . 413
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
Designing a Low-Cost Spin-Drying Desiccation Technique Using 3D Printed Whirligig Model for Preservation of Human Umbilical Cord Blood-Derived Mesenchymal Stem Cells . . . . . . . . . . . . . . . . . . . . . . . . Sharda Gupta, Akalabya Bissoyi, Pradeep Kumar Patra, and Arindam Bit 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Materials and Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Results and Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . .
414 416 416 416 418 418 418 419 423 423
. 425 . . . . .
426 427 430 433 435
Contents
Fabrication of Hydroxyapatite-Chitosan-Silk Fibroin Based Composite Film as Bone Tissue Regeneration Material . . . . Sharda Gupta, Rupsha Mukherjee, Rajendra Kumar Jangle, Deependra Singh, Manju Singh, and Arindam Bit 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Materials and Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Results and Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1 Mechanical Testing . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2 Swelling Ratio . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3 Fourier Transform Infrared Spectroscopy . . . . . . . . . . 3.4 Hemocompatibility Test . . . . . . . . . . . . . . . . . . . . . . 4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
xix
. . . . . . . . . 437
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
Prediction of Hydroxyurea Effect on Sickle Cell Anemia Patients Using Machine Learning Method . . . . . . . . . . . . . . . . . . . . . . . . . Bikesh Kumar Singh, Apoorva Ojha, Kshirodra Kumar Bhoi, Akalabya Bissoyi, and Pradeep Kumar Patra 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Materials and Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1 Data Acquisition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 Simulation Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Result and Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1 Features Selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2 Classification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3 Accuracy, Sensitivity, Specificity, and ROC Curve . . . . . . . 4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
437 439 441 441 442 442 443 444 444
. . . . . 447
. . . . . . . . . .
. . . . . . . . . .
A Survey on IoT-Based Healthcare System: Potential Applications, Issues, and Challenges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Kavita Jaiswal and Veena Anand 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Issues and Challenges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 IoT Healthcare Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Real-Time Data Augmentation Based Transfer Learning Model for Breast Cancer Diagnosis Using Histopathological Images . . . Rishi Rai and Dilip Singh Sisodia 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Related Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Transfer Learning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
447 449 450 450 451 451 452 452 454 456
. . . 459 . . . . .
. . . . .
. . . . .
460 463 467 469 469
. . . . . . 473 . . . . . . 474 . . . . . . 475 . . . . . . 476
xx
4 Methodology . . . . . . . . . . . . . . . . . . . . 4.1 Preprocessing . . . . . . . . . . . . . . . . 4.2 Data Augmentation . . . . . . . . . . . . 5 Models Used for Training . . . . . . . . . . . 5.1 InceptionV3 . . . . . . . . . . . . . . . . . 5.2 Xception Model . . . . . . . . . . . . . . 5.3 3-Layer CNN Model . . . . . . . . . . . 6 Evaluation Parameters . . . . . . . . . . . . . . 6.1 Image Recognition Rate/Accuracy . 6.2 Sensitivity . . . . . . . . . . . . . . . . . . 6.3 Specificity . . . . . . . . . . . . . . . . . . 6.4 Area Under the Curve . . . . . . . . . . 7 Experimental Results . . . . . . . . . . . . . . 7.1 Training Strategies . . . . . . . . . . . . 7.2 Training by Transfer of Learning . . 7.3 Training from Scratch . . . . . . . . . . 7.4 Image Recognition Rate/Accuracy . 7.5 ROC Curve . . . . . . . . . . . . . . . . . 7.6 Sensitivity and Specificity . . . . . . . 7.7 Comparison with Previous Results . 8 Conclusion and Future Work . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . .
Contents
. . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . .
477 477 477 479 479 480 480 480 480 481 482 482 482 482 482 483 483 484 485 486 486 487
About the Editors
Albert A. Rizvanov (Ph.D., Dr. Sci.) graduated from Kazan State University, Russia (biology, microbiology) in 1996. After completing his Ph.D. (2003) in cellular and molecular biology at the University of Nevada, Reno, USA, he undertook his Dr. Sci. (2011) in biochemistry (Habilitation) at Kazan Federal University (KFU), Russia. Currently, Albert Rizvanov is a Professor and Director of the Center for Precision and Regenerative Medicine, Institute of Fundamental Medicine and Biology, KFU. He is the head of the Open Lab Gene and Cell Technologies Laboratory, Director of the Department of Exploratory Researches of Pharmaceutical Research and Education Center and head of the Center of Excellence “Regenerative Medicine”. Additionally, he is the Vice-Director of the Strategic Academic Unit “Translational 7P Medicine” as part of the government program of competitive growth (“5–100 Program”) and the corresponding member of the Tatarstan Academy of Sciences, Russian Federation. Albert is an author on more than 300 peer-reviewed journal articles, 3 book chapters, and 22 patents, he has successfully supervised 15 Ph.D. and 2 Dr. Sci. dissertations, and is the head of the biochemistry, microbiology, and genetics dissertation committee at KFU. He is the principal investigator of more than 50 grants supported by NATO, British Council, Russian Science Foundation, Russian Foundation for Basic Research and other Russian government federal programs and industry contracts. His fields of expertise include regenerative medicine, precision medicine, gene and cell therapy, molecular neurobiology, molecular virology, cancer diagnostics and therapy. In 2019 Albert Rizvanov became an Honorary Professor of Fundamental Medicine at the Faculty of Medicine and Health Sciences, University of Nottingham, UK. Bikesh Kumar Singh (Ph.D.) is Assistant Professor in Department of Biomedical Engineering at National Institute of Technology Raipur, Raipur (Chhattisgarh) India. He obtained his B.E. (Electronics and Telecommunication Engineering) Gold Medalist and M.Tech. (Electronics and Telecommunication Engineering) Honors from Pt. Ravishankar Shukla University, Raipur. He received his Ph.D. in Biomedical Engineering from National Institute of Technology Raipur, Raipur (Chhattisgarh) India. He has published more than 70 research papers in various xxi
xxii
About the Editors
international and national journals and conferences. He is active reviewer and has reviewed several research articles of reputed International Journals. He has teaching and research experience of 12 years. He has been Head of the Department of Department of Biomedical Engineering for 5 years. He is member of International Professional Societies such as IEEE (Senior member) & IACSIT and also of many National Professional bodies like CSI India, IETE India, ISCA India and IEI India. He has received several awards like Chhattisgarh Young Scientist Award, IETE Gowri Memorial Award, IEI Young Engineer Award. He has delivered several expert talks in the area of Machine Learning Applications. He has organized several workshops and international conference in area of Biomedical Engineering, Machine Learning and Softcomputing. His research interest includes applications of machine learning and artificial intelligence in medical image analysis, biomedical signal analysis, computer aided diagnosis, computer vision and cognitive science. Padma Ganasala (Ph.D.) is currently working as Associate Professor in the Department of Electronics and Communications Engineering, Gayatri Vidya Parishad College of Engineering, Visakhapatnam, India. She had received her Ph.D. in Medical Image Fusion from Indian Institute of Technology Roorkee (IIT-ROORKEE). She is a recipient of MHRD fellowship during studies. To her credit, she possesses several publications in reputed international journals and conferences. She has reviewed many journal papers published by prestigious journals and conferences. She is a Life Member of ISTE. Her research interests include medical image processing and analysis, biomedical signal processing, machine learning and deep learning.
Extraction and Phytochemical Analysis of Coccinia indica Fruit Using UV-VIS and FTIR Spectroscopy Alok Sharma, Bidyut Mazumdar, and Amit Keshav
Abstract The phytochemical analysis of Coccinia indica fruit extract was performed employing UV-VIS and FTIR Spectroscopy. Extractions of phytochemicals were carried out using different solvents, selected on the basis of polarity viz. ethanol, methanol, and chloroform. Antioxidant activity of extracts was measured by the DPPH method where ascorbic acid was used as standard. The UV-VIS spectroscopy revealed the characteristic peaks for different phytochemicals present in the extract. The FTIR analysis helped to identify the presence of different functional groups which ultimately leads to the confirmation of existence of phytochemicals in the extract. The phytochemicals thus extracted and identified have major applications in biotechnology, food processing, and pharmaceutical industries. Keywords Coccinia indica · Phytochemicals · FTIR spectroscopy · UV-VIS spectroscopy · DPPH
1 Introduction Coccinia indica is a perennial and creeping plant belonging to Cucurbitaceae family. It is also known as Ivy Gourd, Kundru (Chhattisgarhi), Kovakai (Tamil), Tindora. It is found extensively throughout the Indian Subcontinent. The fruit is berry shaped, green when unripe, and becomes orange-red when ripens. The fruits of this plant are the main constituent of the regular meal in Indian culinary. It has been looked forward for its herbal and medicinal properties as mentioned in Ayurveda. C. indica is also known for its antidiabetic, anti-obesity, antimicrobial, antifungal, antileishmanic, antioxidant, antihypertensive, antitussive, antiulcer, analgesic, antipyretic, antianaphylactic, and anticancer properties (Sakharkar and Chauhan 2017; Singha et al. 2007). The fruits are believed to have numerous phytochemicals which include antioxidants, phenolic compounds, flavonoids, alkaloids, terpenoid, tannins, A. Sharma (B) · B. Mazumdar · A. Keshav Department of Chemical Engineering, NIT Raipur, Raipur, India e-mail: [email protected] © Springer Nature Singapore Pte Ltd. 2021 A. A. Rizvanov et al. (eds.), Advances in Biomedical Engineering and Technology, Lecture Notes in Bioengineering, https://doi.org/10.1007/978-981-15-6329-4_1
1
2
A. Sharma et al.
saponins, glycosides, etc. (Kumar et al. 2014). This current research was focused on exploring the phytochemicals present in C. indica using different analysis methods.
2 Material and Methods Fresh fruits of C. indica were purchased from the local market of Raipur Chhattisgarh (India). It was properly rinsed with water to remove the physical impurities from the surface. Afterwards, the fruits were chopped and sliced by a kitchen slicer to get the uniform slices prior to sun drying. The sliced fruits were subjected to sun drying for 2 days to remove the indigenous water from it. The dried fruit was ground in a mixergrinder. Solvent extraction was studied using ethanol, methanol, chloroform which are of analytical grade and obtained from Merck, India. Ltd. Antioxidant activity was measured using 1, 1-diphenyl-2-picryl-hydrazyl (DPPH) obtained from Merck, India. Ltd. Solvent extraction of phytochemicals was done using the 250 ml Soxhlet apparatus for 12 h using ethanol, methanol, and chloroform as solvents, respectively. About 25 g of dried fruit powder was taken into the thimble to perform the extraction. The extract thus obtained was filtered (Whatman Filter Paper No. 1) and concentrated to get the crude extract, which was then stored at refrigerated temperature until further use. The extracts were analyzed for the identification and characterization of the phytochemicals by UV-VIS Spectrophotometer (Shimadzu UV-1800) and FTIR (Bruker). The extracts were diluted to the extent of 1:10 for the same solvent prior to UVVIS Spectrometry. FTIR Spectroscopy was done for identification of the functional groups present in the extract. The KBr thin disc was formed for this analysis, which was made by mixing small amount of C. indica extract with dry potassium bromide. Further, the disc was placed over the sample cup of diffuse reflectance accessory. The extracts were analyzed using FTIR Spectrophotometer where IR spectrum was within 4000–400 cm−1 . The results thus obtained from UV-VIS and FTIR were recorded. The antioxidant activity was evaluated by DPPH assay as mentioned by Baba and Malik (Baba and Malik 2015). About 3.8 ml of freshly prepared DPPH solution was taken and 200μL of extract was added for each solvent viz. ethanol, methanol, and chloroform. Further the reaction mixture was incubated in dark for 1 h at room temperature. The measurement of absorbance was done at 517 nm wavelength by UV-VIS spectrometer. Ascorbic acid was used as positive control. The DPPH activity was calculated by the formula mentioned below: DPPH Activity =
Contr ol absor bance − Sample absor bance × 100 Contr ol absor bance
Extraction and Phytochemical Analysis of Coccinia indica Fruit …
3
3 Result and Discussion The extraction of phytochemicals from C. indica fruits was found to be more effective for chloroform than ethanol and methanol. The UV-VIS Spectroscopy shows the different absorbance peaks obtained for the respective phytochemical which is also mentioned in literatures. The UV-Vis spectroscopy dictates that the phytochemicals are present in the extract based on the λmax values for each phytochemical. Table 1 depicts the absorbance values (λmax ) of phytochemicals with respect to the wavelength, for ethanol, methanol, and chloroform, respectively. Chlorophyll a, Chlorophyll b, Taraxerol, β amyrin, Lupeol, and Sitosterol were the phytochemicals detected in the ethanolic and methanolic extracts of C. indica fruit (Wang et al. 2007; Laphookhieo 2004; El-Alfy et al. 2011; Quilitzsch et al. 2005; Jain and Bari 2010; Chung and Hahn 2005). Similar phytochemicals were detected in the chloroform extract. β carotene was only detected in the extracts of chloroform. The λmax identified phytochemicals was mentioned in the literatures (Okoye and Daniel 2014; Khanra et al. 2014; Miller et al. 1936; Mallick 2014). The FTIR spectroscopy was used to identify the functional groups of phytochemicals present in the extracts of C. indica fruit. The peak values obtained by the FTIR spectroscopy validate the presence of particular functional groups that a particular phytochemical contains. The results of FTIR peak values and functional groups have been illustrated in Table 2 and Figs. 1, 2 and 3. The FTIR spectrum profile confirms the presence of functional groups in ethanol, methanol, and chloroform extracts of C. indica. Amides were detected at 3325, 3326, and 2839 cm−1 wavenumber in ethanol and methanol extracts. Alkanes were detected in ethanol and chloroform extracts at 2923, 2924, and 2852 cm−1 wavenumber. Alkene was observed in methanol and chloroform extracts at 1646 and 1658 cm−1 . 1° amine was detected at 1653 and 1631 cm−1 in ethanol and chloroform extracts. Aromatics were identified in all the solvents at 1043, 868, 1447, 1406, 1462, 772, and 686 cm−1 . Nitro groups were detected in extracts of ethanol and methanol at 1378 and 1554 cm−1 . Table 1 Phytochemicals identified in different extracts by UV-Vis spectroscopy S. No
Phytochemical
Ethanol extract
Methanol extract
Chloroform extract
λmax
Abs
λmax
Abs
λmax
Abs
1
Chlorophyll b
665
0.485
665
0.495
665
0.617
2
Chlorophyll a
606
0.150
606
0.170
607
0.203
3
Taraxerol
535
0.234
535
0.230
536
0.261
4
β amyrin
409
2.188
409
2.158
436
1.317
5
β carotene
6
Lupeol
314
4.0
314
7
Sitosterol
269
4.0
269
412
1.978
3.96
322
1.364
3.984
268
1.495
4
A. Sharma et al.
Table 2 FTIR peak values of identified functional groups in different extracts Ethanol
Methanol
Wave number cm−1
Functional group
wave number cm−1
3325
Amides
2969
Alkanes
2923 1653
Chloroform Functional group
wave number cm−1
Functional group
3326
Amides
3425
Phenols
2839
Amides
2924
Alkanes
Alkanes
1646
Alkene
2852
Alkanes
1°amines
1447
Aromatic
1742
Esters
1378
Nitro compound
1406
Aromatic
1658
Alkene
1085
Aliphatic Amines
1017
Aliphatic Amine
1631
1°amine
1043
Aliphatic Amines
541
Alkyl halide
1554
Nitro compound
878
Aromatic
518
Alkyl halide
1536
Unknown
525
Alkyl Halides
510
Alkyl halide
1462
Aromatic
514
Alkyl Halides
1382
Alkane
1219
Aliphatic Amine
1032
Aliphatic Amine
772
Aromatic
686
Aromatic
The wavenumber 1085, 1017, 1219, and 1032 cm−1 denoted the presence of aliphatic amine in all the extracts. Alkyl halides were present in ethanol and methanol extracts at the wavenumber of 525, 514, 518, and 510 cm−1 . Moreover, phenols and esters were identified only at 3425 and 1742 cm−1 respectively in chloroform extracts only. The functional groups thus detected in the different extracts leads to the assumption that a variety of phytochemicals are present in the C. indica fruits. DPPH test is useful to determine the radical scavenging activity of extraction. The method relies on the decrease in absorption of DPPH solution after addition of antioxidant. The standard for this test is done using ascorbic acid. DPPH has red color and degree of discoloration indicates the scavenging potential of the antioxidant. DPPH radical scavenging activity was calculated using the absorbance values obtained by spectrophotometer. The ethanol extracts showed 60.15% of activity, while it was 60.46% and 61.07% for methanol and chloroform, respectively.
Extraction and Phytochemical Analysis of Coccinia indica Fruit …
Fig. 1 FTIR spectra of ethanolic extract of C. indica fruit
Fig. 2 FTIR spectra of methanolic extract of C. indica fruit
5
6
A. Sharma et al.
Fig. 3 FTIR spectroscopy of chloroform extract of C. indica fruit
4 Conclusion This research depicts that UV-VIS and FTIR Spectroscopy can be applied for the phytochemical analysis of C. indica fruit extracts. Methanol extracts proved to be better than ethanol and chloroform in terms of DPPH radical scavenging activity. Some more techniques can also be used for better profiling of phytochemicals.
References Baba SA, Malik SA (2015) Determination of phenolic and flavonoid content, antimicrobial and antioxidant activity of a root extract of Arisaema jacquemontii. J Taibah Univ Sci 9:449–454 Chung I-m, Hahn S-j, Ahmad A (2005) Confirmation of potential herbicidal agents in hulls of rice Oryza sativa. J Chem Ecol 31(6):1339–1352 El-Alfy TS, Ezzat SM, Hegazy AK, Amer AMM, Kamel GM (2011) Isolation of biologically active constituents from Moringa peregrina (Forssk.) Fiori. (family: Moringaceae) growing in Egypt. Pharma Magazine 7(26):109–115 Jain PS, Bari SD (2010) Isolation of Lupeol, Stigmasterol and Campesterol from petroleum ether extract of woody stem of Wrightia tinctoria. Asian J Plant Sci 9(3):163–167 Khanra R, Dewanjee S, Dua TK, Sahu R, Gangopadhyay M, De Feo V, Zia-Ul-Haq M (2014) Abroma augusta L. (Malvaceae) leaf extract attenuates diabetes induced nephropathy and cardiomyopathy via inhibition of oxidative stress and inflammatory response. J Trans Med 13:1–14 Kumar M, Alok S, Jain SK, Dixit VK (2014) Macroscopial, anatomical and physico-chemical studies on fruits of Coccinia indica wight and Arn. (Cucurbitaceae). Asian Pac J Trop Dis 4(1):S121–S128
Extraction and Phytochemical Analysis of Coccinia indica Fruit …
7
Laphookhieo S, Karalai C, Ponglimanont C (2004) New Sesquiterpenoid and Triterpenoids from the Fruits of Rhizophora mucronate. Chem Pharm Bull 52(7):883–885 Mallick SS, Dighe VV (2014) Detection and estimation of alpha-Amyrin, beta-Sitosterol, Lupeol, and n-Triacontane in two medicinal plants by high performance thin layer chromatography. Adv Chem 1–7 Miller ES, Inney GM, Zscheile FP Jr. (1936) Absorption spectra of alpha and beta carotenes and lycopene. Plant Physiol. 9(1):375–381 Okoye NN, Ajaghaku DL, Okeke HN, Ilodigwe EE, Nworu CS, Okoye FBC (2014) Beta-Amyrin and alpha-amyrin acetate isolated from the stem bark of Alstonia boonei display profound antiinflammatory activity. Informa Healthcare 52(11):1478–1486 Quilitzsch R, Baranska M, Schulz H, Hoberg E (2005) Fast determination of carrot quality by spectroscopy methods in the UV-VIS, NIR and IR range. J Appl Bot Food Qual 79:163–167 Sakharkar P, Chauhan BS (2017) Antibacterial, antioxidant and cell proliferative properties of Coccinia grandis fruits. Avicenna J Phytomed 7(4):295–307 Singha G, Gupta P, Rawat P, Purib A, Bhatiab G, Maurya R (2007) Antidyslipidemic activity of polyprenol from Coccinia grandis in high-fat diet-fed hamster model. Phytomedicine 14:792–798 Wang L, Zhang C, Wu F, Deng N (2007) Photodegradation of aniline in aqueous suspensions of microalgae. J Photochem Photobiol B Biol 87:49–57
Numerical Simulation Method to Predict Air Flow and Contaminant Control in a Multiple Bed Intensive Care Unit of Hospital Arvind Kumar Sahu, Shobha Lata Sinha, and Tikendra Nath Verma
Abstract A hospital’s physical pattern is a vital part of its contamination management measures to reduce the danger of spread of any communicable disease. Recent and rising communicable diseases challenges as higher public prospects and awareness of care connected problems, a lot of thought has got to incline to the layout of the hospital. In present study, airflow simulation of a multi-patients ICU room has been carried out using FLUENT version 15 CFD software. For simulation of airflow standard k-epsilon turbulence model used with high-quality unstructured mesh. Five different cases of multiple staff orientation have been studied to look at the infection between every patients and additionally for medical staff. In whole cases, inlet fresh airflow temperature (273 K) and airflow rate (0.2 m/s) are held constant. An average of 9.78 min time was taken by mobile contaminants to leave ICU room. The location of air inlet and outlet holds good air ventilation as particles coming out from the mouth of patient moves out of ICU in most of the instances and hospital staff orientations. Keywords Room airflow · Buoyancy · Recirculation zone · k-epsilon model · Room airflow · Breathing
1 Introduction Intensive care units of hospital require excellent medical supervision staff i.e., physician and nurses for well caring of critically ill patients. ICU of hospitals is commonly proposed for multi patients, these ICUs or wards are economically well-organized in which critically ill patients can be placed along one hall or in two rows, which facilitates less time for supervision. Airflow analysis of multiple patient wards or ICU room’s ventilation plays a vital role on health of patients and medical staffs as well. In the present investigation, numerical simulation of ICU room for multiple patients has been carried out. The objective of current study is to envision the effect A. K. Sahu (B) · S. L. Sinha · T. N. Verma Department of Mechanical Engineering, NIT Raipur, Raipur, India e-mail: [email protected] © Springer Nature Singapore Pte Ltd. 2021 A. A. Rizvanov et al. (eds.), Advances in Biomedical Engineering and Technology, Lecture Notes in Bioengineering, https://doi.org/10.1007/978-981-15-6329-4_2
9
10
A. K. Sahu et al.
of contaminant which is getting out from the mouth of patients on different orientations of hospital staff to recover out the safest orientation to prevent hospital staffs and patients from infection. A total of 5 different cases of two staff orientations have been studied to examine the infection between each patient and also for the staff. For all cases, airflow rate and temperature of inlet air are kept constant. Various case studies using CFD for fluid flow in ventilations of multiple/single bed hospitals were performed. Contaminant distribution considered with variable air volume (VAV) and heating, ventilation and air conditioning (HVAC) and systems in hospitals. The effect of the location of the infected patients in transmitting air-borne diseases with hospitals using CFD is also presented (Sinha et al. 2000; Prakash and Ravi Kumar 2015; Verma and Sinha 2015a,b; Senthilkumar and Raju 2016). It is found that CFD applications are beneficial to assist and recognize the suitability of the ventilation system in renewal of hospital design to fulfill the latest engineering standards (Chow and Yang 2003). An additional study also revealed that the respiratory events such as breathing, sneezing, talking, and coughing were the main source of transfer of contaminants. The study uses a model of equations using various parameters like rate of flow, area of mouth orifice opening, etc., and proposed that the model be used for describing source of contaminant transfer due to talking and breathing (Gupta et al. 2010). The airflow patterns of an operating room (OR) during opening and foot traffic are studied. Even though OR has slightly higher pressure than other adjacent rooms, a small volume of air still enters during a cycle of door opening and closing even without any person entering the room. The study has revealed a higher volume of air enters the OR if the person enters the OR (Villafruela et al. 2013). The evaporation and condensation of expiration droplets and their size (coughing and speaking) have been found to have negligible impact on usual droplet size from human beings (13.5 μm from coughing and 16.0 μm from speaking) for average expiration velocity of 11.7 m/s and 3.9 m/s for coughing and speaking, respectively (Chao et al. 2009). Contaminant distribution in an office environment of 6.6 m (L) × 3.7 m (W) × 2.6 m (H) dimension with air conditioning and mechanical ventilation was also studied. Tracer gas (SF6 ) is used for simulation of contaminants on a model room and CFD was used for validating the results. The study revealed that the pattern of contaminant dispersion depends greatly on the velocity flow field. The layout of various objects like furniture also influences the pattern of airflow and contaminant. Another study suggests that CFD can be used effectively in predicting the spatial distribution of bio-aerosol in indoor environments like hospitals. The study was conducted at three different layouts—empty, single bedroom, and two bedrooms. Deposition of the particles has no correlation with relative surface concentration and source distance but partition among the patients proved to be effective in reducing cross contamination among patients. (Cheong et al. 2003; King et al. 2013). Authors have performed various experiments in different regions using the respective environment conditions. The authors in this paper have performed real time study by means of experimental and numerical solution using environment conditions of the hospital in Raipur (21.2514° N, 81.6296° E, 298 m altitude above sea level), Chhattisgarh (Sahu 2018; Verma 2018,2014; Verma and Sinha 2013).
Numerical Simulation Method to Predict Air Flow …
11
2 Governing Equations To study air spread pattern inside ICU model room FLUENT version 15 CFD tool has been used. GAMBIT preprocessing software is used to create high-quality unstructured grids. Grid independence test carried out before simulation and after GIT computation domain has 5.2 × 106 grids. Turbulence in airflow is considered at the entry of ICU model room and K-epsilon turbulence model is used for processing airflow in computational domain. Here airflow in an ICU model room is described by mass, momentum, and energy conversion equations which are basically quantity of fluid per second, newton’s second law of motion, and thermodynamics first law, respectively. Mathematically it is expressed as (Patankar 1980). ∂ (ρφ) + div(ρuφ) = div Tφ gradφ + Sφ ∂t
(1)
Motion of mobile contaminant particles are tracked using newton’s second law of motion (Patankar 1980). ∂up 18μCd Re = Fi , FD = ∂t 24ρp dp2
(2)
g(ρp − ρ) dup + Fx = FD ui − up + dt ρp
(3)
where Fi represents external forces exerted on the particle and FD represents drag force (N), Cd is coefficient of drag, ρ is Mass density in kg/m3 , μ is dynamic viscosity in Ns/m2 and Fx is additional forces exerted on particles. P denotes for particle and i is Particle identifier. The above equation is integrated with particle tracking module of FLUENT software and used for tracking mobile contaminant. The following assumptions have been used during computation: • • • • • • •
The cross section of beds are considered rectangular; Walls, floors, and roof of ICU model room are considered well insulated; Lying position of patients is the east–west direction; At a time only one patient is considered to be producing mobile contaminant; Shape of mobile contaminant particle is considered solid spherical; No contaminants recoil on solid walls; Motion of one mobile contaminant is considered for clarity of particle tracking in figures; • Heat and mass transmission between mobile contaminant and air inside ICU room is neglected; • Diameter of mobile contaminant is assumed to be circular and uniform.
12
A. K. Sahu et al.
3 Problem Statement Figure 1 shows the outline of five patient ICU model room with 5 different orientation of hospital staff. The outline of ICU room has two rectangular inlet vents and two outlet vents for entrance of fresh air and elimination of sick air from the model ICU room. The ICU model room is inspired by actual ICU room of a hospital which is situated at BALCO region of Korba C.G. position of inlet, outlet vents, and lying arrangement of ICU model room is considered same as actual. Height, width, and length of ICU model room are 3.0 m, 5.8 m, and 6.3 m, respectively and sizes of all inlet and outlet vents are 0.6 m × 0.4 m. The location of both inlet vents is kept constant i.e., 2.3 m overhead the ground, inlet 1 is 1.4 m and inlet 2 is 3.8 m ahead the east wall. The positioning of staff is based on literature surveys and surveys of various hospitals of Chhattisgarh. Inlet airflow rate in ICU room is considered 0.2 m/s. The temperature of fresh air is considered 293 K for all five cases and air properties are taken as per this temperature i.e., 293 K (Table 1). Temperature of different walls has been selected from the ISHRAE handbook2007 for Raipur (C.G.) region (Table 2).
4 Result and Discussion Figure 2a, b shows the variation in velocity vector for two staff orientation case 1, where the temperature of fresh air stream is considered as 293 K, inlet air velocity is considered 0.2 m/s. The fresh and unpolluted air stream enters from inlet vents and throws well into ICU room. After throwing it mixes and spreads well with the existing air of ICU room. After appropriate distribution in ICU room, the air mixture drops near to opposite walls and moves out from the outlet vents. Figure 2a shows airflow pattern on plane 1.7 m ahead of east wall, it is clear from Fig. 2a that one recirculation zone is formed near south wall. Figure 2b shows variation in velocity vector on plane 4.1 m ahead of east wall, it is clear from the Fig. 2b one recirculation zone is found between patient 2 and patient 3 and smaller air circulation zone found at right corner near to north wall. Figure 3a, b shows the temperature contour on the plane 1.7 and 4.1 m ahead of east wall for two hospital staff orientation case 1.Colors in Fig. 4a–e show time for contaminant to leave ICU in minutes. Here red color stands for maximum and blue stands for minimum contaminant leaving time. The temperature of the roof and different walls are considered different due to the difference in incident solar radiation, which is clear from Fig. 3a, b. Mixing of fresh air with existing air is found appropriate and uniform at occupied zone. Temperature boundary layers are formed near to walls of ICU room due to temperature variation in walls (Table 1). Also small temperature variation due to airflow rate is found at the entry of airflow and right corner of the ICU room.
Numerical Simulation Method to Predict Air Flow …
13
(a) Hospital staff orientation 1
(b) Hospital staff orientation 2
(c) Hospital staff orientation 3
(d) Hospital staff orientation 4
(e) Hospital staff orientation 5 Fig. 1 Five-bed ICU room with different hospital staff orientations
14
A. K. Sahu et al.
Table 1 Air properties at atmospheric pressure Flow type
Cold
Temp
Density
Specific heat
Thermal conductivity
Dynamic viscosity
(K)
(kg/m3 )
(J/kg-K)
(W/m-K)
(kg/m-s)
293
1.204
1007
0.02514
1.83 × 10–05
Table 2 Temperature of walls Walls
North
South
East
West
Celling
Floor
Temp (K)
290
301
304
298
321
296
INLETS
OUTLETS Colored by velocity magnitude in m/s
(a) Velocity vector at plane x=1.7 m (inlet air velocity 0.2 m/s)
INLETS
OUTLETS Colored by velocity magnitude in m/s
(b) Velocity vector at plane x=4.1 m (inlet air velocity 0.2 m/s)
Fig. 2 a, b Velocity vector for two hospital staff orientation 1 (ACH = 2.05, Re = 12590)
Numerical Simulation Method to Predict Air Flow …
15
INLETS
OUTLETS Colored by temperature magnitude in °K
(a) Temperature contour at plane x=1.7 m (inlet air velocity 0.2 m/s)
INLETS
OUTLETS Colored by temperature magnitude in °K
(b) Temperature contour at plane x=4.1 m (inlet air velocity 0.2 m/s)
Fig. 3 a, b Temperature contour for two hospital staff orientation 1 (ACH = 2.05, Re = 12590)
Figure 4a–e shows the movement of mobile contaminant for two staff orientation case 1, for different particle which emergences from the mouth of ill patients in the ICU room. Here, the profile of motion of the contaminated particles which is coming out from patients are not affecting other patients and occupants near the patients. Table 3 shows the time required for elimination of mobile contaminant from ICU room. It is clear from Table 3 that elimination time for the mobile contaminant from patient 1 is comparatively less than other patients.
16
A. K. Sahu et al.
Fig. 4 a–e Movement of mobile contaminant through patients for hospital staff orientation 1 (ACH = 2.05, Re = 12590)
0.00 8.58 Colored by time in minute (a) Movement of mobile contaminant through patient 1
0.00 5.08 Colored by time in minute (b) Movement of mobile contaminant through patient 2
0.00 6.30 Colored by time in minute (c) Movement of mobile contaminant through patient 3
Numerical Simulation Method to Predict Air Flow …
17
Fig. 4 (continued)
0.00 4.80 Colored by time in minute (d) Movement of mobile contaminant through patient 4
0.00 6.17 Colored by time in minute (e) Movement of mobile contaminant through patient 5
5 Conclusion The numerical simulations of airflow have been carried out for ICU room. Control volume method is used to solve energy and Navier Stokes equations by utilizing FLUENT software. It is a realistic flow problem which brings into account the outcome of multiple patients and multiple staff as obstacles in a room. It is seen that ventilation execution is fully influenced upon airflow rate and its entry and exit
18
A. K. Sahu et al.
Table 3 Mobile contaminant elimination time (ACH = 2.05, Re = 12590) Case No
Time in minute Patient 1
Patient 2
Patient 3
Patient 4
Patient 5
1
8.58
7.62
19.50
5.13
2
5.08
7.78
11.87
6.88
5.13 4.75
3
6.30
6.97
11.05
16.67
28.50
4
4.80
20.67
6.42
9.60
7.08
5
6.17
7.77
17.50
5.70
6.90
locations of hospital. The fresh airflow rate in ICU should be a sufficient amount and spread rate should be capable to eliminate mobile contaminants from all portion of the hospital/isolation rooms. The air spread rate ought to be inside with worthy limits of commotion level and human consolation. Standard k-epsilon turbulence model is rigidly valid for turbulent flows. In this investigation, basic dimensions like length, height, and width of ICU of hospital are kept constant. To compare effect of various orientation of hospital staff, the location of inlet vents, outlet vents, and beds along with ill patients are kept constant for all cases. Average time for elimination of mobile contaminant is found as 9.78 min. The location of air inlet and outlet holds great air ventilation as particles coming out from the mouth of patients move out of ICU in most occasions and hospital staff orientations. By perception, it is prompted to keep location of the nurture-station and patient caring staff closed to inlet vents to keep them disease free.
References Chao CYH, Wan MP, Morawska L, Johnson GR, Ristovski ZD, Hargreaves M, Mengersen K, Corbett S, Li Y, Xie X, Katoshevski D (2009) characterization of expiration air jets and droplet size distributions immediately at the mouth opening. J Aerosol Sci 40:122–133 Cheong KWD, Djunaedy E, Poh TK, Tham KW, Sekhar SC, Wong NH, Ulah MB (2003) Measurements and computations of contaminant’s distribution in an office environment. Build Environ 38:135–145 Chow TT, Yang XY (2003) Performance of ventilation system in a nonstandard operating room. Build Environ 38(12):1401–1411 Gupta JK, Lin CH, Chen Q (2010) Characterizing exhaled airflow from breathing and talking. Indoor Air 20:31–39 King MF, Noakes CJ, Sleigh PA, Camargo-Valero MA (2013) Bio-aerosol deposition in single and two-bed hospital rooms, a numerical and experimental study. Build Environ 59:436–447 Patankar SV (1980) Numerical heat transfer and fluid flow. McGraw Hill, Washington Prakash D, Ravi Kumar P (2015) Analysis of thermal comfort and indoor air flow characteristics for a residential building room under generalized window opening position at the adjacent walls. Int J Sustain Built Environ 4:42–57 Sahu AK, Sinha SL, Verma TN (2018) Numerical simulation of air flow to ventilate intensive care unit of hospital, computer application in education & research for science and technology. Int Res Public House Delhi 131–138
Numerical Simulation Method to Predict Air Flow …
19
Senthilkumar M, Raju NMS (2016) transient effect of window position on naturally ventilated room with various vent cross section area. Int J Adv Eng Technol Sinha SL, Arora RC, Roy S (2000) Numerical simulation of two dimensional room air flow with and without buoyancy. Energy Build 32(1):121–129 Verma TN, Sinha SL (2013) Contaminant control in intensive care unit (ICU) using CFD modeling. Int J Mechan Industr Eng 3:121–125 Verma TN, Sinha ST (2014) Contaminant control in intensive care unit of hospital. Appl Mechan Mater 2486–2490 Verma TN, Sinha SL (2015) Trajectory of contaminated particle in intensive care unit of hospitals using numerical modelling. Int J Des Manuf Technol 9(1) Verma TN, Sinha SL (2015) Numerical simulation of contaminant control in multi-patient intensive care unit of hospital using computational fluid dynamics. J Med Imaging Health Inform 5:1–5 Verma TN, Sahu AK, Sinha ST (2018) Numerical simulation of air pollution control in hospital. Energy Environ Sustain 185–206 Villafruela JM, Olmedo I, Ruiz M, de Adana, Méndez C, Nielsen PV (2013) CFD analysis of the human exhalation flow using different boundary conditions and ventilation strategies. Build Environ 62:191–200
A Simple Robust Image Processing Algorithm for Analysis of Static Foot Pressure Intensity Image to Detect Foot Risk Areas in Diabetic Patients Hari S. Nair, Navya Thomas, and R. Periyasamy
Abstract According to WHO, about 422 million people are affected with diabetics and among them 15% of diabetic patients are associated with foot problems. Due to peripheral neuropathy, the patients are unable to sense pressure pain which eventually creates calluses. Thus the aim of this study is to effectively find the risk areas which may form ulcers in the foot. The physical components required for this study are a camera which captures high-quality images of foot and a laptop capable of processing image by using MATLAB (The Mathworks Inc., USA). The algorithm which we developed extracts the high-intensity areas in the foot which are actually the areas where the patient exert high pressure compared to other areas in the foot. This method was evaluated by assessing the foot for 5 normal patients and 5 diabetic patients. It is found that this method is capable of giving the riskier areas of the foot and thereby further prevention of foot ulcer can be done. Our method is time-efficient, cost-efficient, and space-efficient and thus can be easily installed in any hospital for diabetic foot analysis. In most of the cases, the metatarsal head region and plantar medial tubercle region are mainly affected by calluses. By proper diagnosis and daily care of the foot can reduce the risk of ulcer in diabetic patients. Keywords Callus · Ulcers · Metatarsal head region · Plantar tubercle region
Abbreviations WHO World Health Organization USA United States of America CPU Central Processing Unit H. S. Nair (B) · N. Thomas Department of Biomedical Engineering, National Institute of Technology Raipur, Raipur, India e-mail: [email protected] R. Periyasamy Department of Instrumentation and Control Engineering, National Institute of Technology Trichy, Trichy, India © Springer Nature Singapore Pte Ltd. 2021 A. A. Rizvanov et al. (eds.), Advances in Biomedical Engineering and Technology, Lecture Notes in Bioengineering, https://doi.org/10.1007/978-981-15-6329-4_3
21
22
H. S. Nair et al.
1 Introduction Diabetes being a serious chronic disease needs global attention. The global prevalence of diabetics among adults over 18 years of age has risen from 4.7% in 1980 to 8.5% in 2014 that is about 422 million people (Mathers and Loncar 2002). About 15% of people affected with diabetes are associated with foot problems (Dayananda and Kiran 2014). Diabetic foot problems such as ulcerations, infections, and gangrene are the major cause of hospitalization in diabetic patients (Ingrid and Steven 2006). Diabetic foot ulcer is a common complication which is a major source of morbidity and it is showing an increasing trend over previous decades (Leila et al. 2015). Since the sensory neurons are affected in diabetes, patients are unable to sense pressure pain and microtrauma in the foot resulting in tissue breakdown and formation of cavity at the level of epithelial layer (Prabhu et al. 2001). Peripheral neuropathy and arterial occlusion are mainly the causes of diabetic foot ulcers. Diabetic neuropathy will develop in 50% of type1 and type 2 patients with diabetes (Hajieh et al. 2013). The foot ulcer once developed leads to an increased risk ulcer progression that may lead to amputation. Studies show that the rate of lower limb amputation in patients with diabetes mellitus is 15 times higher than those without diabetes. The multiple risk factors which are associated with the development of foot ulcer as per recent studies are gender (male), duration of diabetes longer than 10 years, advanced age of patients, high body mass index, and other comorbidities such as retinopathy, diabetic peripheral neuropathy, peripheral vascular disease, glycated hemoglobin level, foot deformity, high plantar pressure, infections, and inappropriate foot self-care habits (Leila et al. 2015). Poorly fitting shoes, poor foot care often cause foot deformation that can lead to the formation of callus. The formation of calluses is clearly been associated with foot ulceration (Pavicic and Korting 2006). More than 70% of patients who have developed diabetic foot ulcer experience an exacerbation of the disease in the next 5 years. The ulcer usually appears in the same extremity or the extremity of the opposite side; at least a quarter of these ulcers do not heal (Bijan et al. 2013). In this study, the image of diabetic foot is captured and by proper processing of the image, the possibility to predict the risk areas on the foot that may eventually develop to a foot ulcer is studied. By proper caring of the risk areas, infection can be prevented in diabetic patients. Thus by proposing a method which can effectively show the risk areas on the foot, foot ulcer formation can be prevented. Previously, plantar pressure is measured in laboratories or in hospitals which require heavy equipment such as pressure platforms and mats. Currently in-shoe pressure measurements are being under research which is less bulky and easy to use as compared to the previous one. But both these devices lack the ability to measure the plantar pressure in real-life situations. These methods will be able to pressure hotspots in case of laboratory situations only. It is also important to note that these techniques need bulky devices which tend to increase the cost of setting up in hospitals and also require more efficient and qualified technicians. Lei Wang et al. (Lei et al. 2015) propose a method for the assessment of diabetic foot ulcers by capturing the RGB image of diabetic foot using a smartphone and the
A Simple Robust Image Processing Algorithm for Analysis …
23
wound assessment algorithms include: (1) mean-shift-based image segmentation, which groups all image pixels into a number of homogeneous regions; (2) a fast method for detecting the largest connected component, to recognize the foot outline; and (3) final wound boundary determination achieved by analyzing the internal and external boundaries of the foot outline. After image processing, a wound healing score ranging from 1 to 10 is given to each subject. This technique can be applied only after the wound has occurred and early wound analysis is not proposed in this method. Dayananda and Kiran (2014) developed an image processing algorithm for monitoring the insole wear patterns. In this study, advanced Gabor filters were evaluated to detect the skin irregularities by using a photograph of the insole captured by a smartphone. This method used fuzzy set clustering for clustering the image. As this method used an ordinary smartphone for image capture which effectively reduces the cost but eventually reduces the image quality and thereby loss of image data may occur. Dmitry et al. (2011) proposed a method to predict the tissues which are at the risk of ulceration by using hyperspectral tissue oximetry. Type1 and type2 diabetes mellitus subjects that are under risk of ulceration undergoes hyperspectral image along tissue oximetry. The data are retroactively analyzed and an ulceration prediction index is developed. They then developed an image processing algorithm based on the ulceration index. This algorithm is then capable to predict the tissue under the risk of ulceration. This method requires complex hardware and also the patients are asked for several visits for the process.
2 Methods 2.1 Foot Image Data Acquisition Foot images are acquired by using a suitable camera that can capture high-quality images. Two images are captured, one by placing the foot and the other without the foot which can be used as a reference image. It is done so that the noises that may creep in can be erased using the reference image. These images can be easily transferred to a computer in which the next step processing takes place. The captured image is then saved using the name of the patient along with the date in which the image is captured for future reference. Clinicians can afterwards use these images to compare the current situation of the patient and can determine the stage of healing of the ulcer. This research protocol is a small part of large study which was carried out in accordance with the guidance of the Institutional Ethical Committee NIT Raipur and all the participants have signed the inform consent.
24
H. S. Nair et al.
Image capture
Preprocessing
Histogram equalization and sharpening
Extracting the high intensity areas
Fig. 1 Flow diagram for the process
2.2 Image Processing Algorithms For processing the images, we are using MATLAB (The Mathworks Inc., USA). During wound image processing, the risk areas are found out by 4 modules: (1) background correction; (2) histogram equalization; (3) sharpening the images; (4) finding the risk areas by extracting the high-intensity areas from the processed image Fig. 1. The captured images first need to be converted to grayscale for further processing. The image is first made to be free from background noises. It can be removed by subtracting the foot images from their respective reference images and filling the holes using “imfill” function available in MATLAB. These pre-processed images need histogram equalization which can be done by using “adapthisteq” function followed by sharpening of the edges by using “imsharpen” function. It is found that the risk areas are of higher intensities than the other parts of foot due to the formation of calluses. Thus the higher intensity regions are extracted from the images. The processed image can be used to conclude the extent of risk for getting ulcer. From the subjects whom we used for data acquisition, it is found out that in case of diabetic foot (having risk of ulcer), the processed image stresses on some part of the image while this phenomenon is absent in the case of normal foot. The processed images of both normal foot and diabetic foot along with the captured images are shown in Figs. 2 and 3.
3 Results The results after processing are shown in Figs. 1 and 2 both in the cases of normal patients and diabetic patients. From Figs. 1 and 2, diabetic patients are under greater risk of foot ulcer. In the case of diabetic patients the processed image highlights some regions in the foot which are confirmed as risk areas. Thus for those patients who are under risk, proper diagnosis and daily care of the foot can prevent ulcer. It is also found that the risk areas are mainly in the first, second, and third metatarsal head and in the plantar medial tubercle region of the foot. In the algorithm, the “limit” can take any value between 0 and 255. But the preferred value must lie between 225 and 245. The output images for different values are shown in Figs. 1 and 4.
A Simple Robust Image Processing Algorithm for Analysis …
25
Fig. 2 Input image, reference image and processed images of normal patients
From Fig. 3, it shows that as the limit value increases, the highlighted area decreases. For limit = 225, the image shows even the low-risk areas while for limit = 235, image shows the areas which needs attention and for limit = 245, the areas present (if any) is of greater risk that is at the verge of ulcer.
4 Conclusion The image processing algorithm which we developed can efficiently give the risk areas in the foot of a diabetic patient. The patient should be educated regarding the foot ulcer and the chance he/she may get affected by it. With proper technologies it is possible to trace the walking pattern of the patient, then he/she can be trained to change his walking pattern so that the pressure is evenly distributed throughout his/her foot. This can prevent foot ulcer to an extent. For those patients who are under risk can be asked for frequent visits so that condition of the foot can be observed regularly by comparing the images with the images of the previous visit.
26
H. S. Nair et al.
Fig. 3 Input images, reference images and processed images of diabetic patients
Fig. 4 Output images for different values of limit
The total computing time for the algorithm implemented on a laptop with CPU (Intel i7 3.5 GHz) is nearly 1.1 to 1.2 s for images in dimension 640 * 480 pixels. Thus this method is very time-efficient for calculating the risk areas. As the required equipment comprises of an infrared camera and a laptop, it is cost-efficient as well as space-efficient compared to other clinical instruments. Thus it is easy to install it in any hospital for treating foot ulcers.
A Simple Robust Image Processing Algorithm for Analysis …
Appendix The image processing algorithm works as follows: clc,clear; %reading the original image and reference image he=imread('input foot image.bmp'); he1=imread('reference image.bmp'); %converƟng to grayscale he=rgb2gray(he); he1=rgb2gray(he1); %background correcƟon he3=he-he1; he4=abs(he3); he5=he4>mean2(he4); b=imfill(he5,'holes'); b=double(b); he=double(he); c=b.*he; c=uint8(c); %histogram equalisaƟon d=adapthisteq(c,'cliplimit',0.05); e=imsharpen(d,'amount',5); %extracƟng hish intensity regions limit=235; f=e>limit; f=f*255; %edge capturing technique BW1 = edge(he3,'prewiƩ'); BW1=BW1*255; g=BW1+double(f); figure,imshow(uint8(g));
27
28
H. S. Nair et al.
References Bijan I, Fariborz K, Alireza E, Gholamreza A (2013) Prevention of diabetic foot ulcer. Int J Prev Med 4(3):373–376 Dayananda KJ, Kiran KP (2014) Analysis of foot sole image using image processing algorithms. In: 2014 IEEE global humanitarian technology conference—South Asia satellite (GHTC-SAS), Trivandrum, pp 57–63 Dmitry Y, Aksone N, Kevin S, Laurent P (2011) Assessing diabetic foot ulcer development risk with hyperspectral tissue oximetry. J Biomed Opt 16(2):026009 Hajieh S, Leila Y, Seyed ML (2013) Risk assessment of patients with diabetes for foot ulcers according to risk classification consensus of international working group on diabetic foot (IWGDF). Pak J Med Sci 29(3):730–734 Ingrid K, Steven E (2006) Evaluation and treatment of diabetic foot ulcer. Clin Diabetes 24(2):91–93 Leila Y, Morteza N, Sara A (2015) Literature review on the management of diabetic foot ulcer. World J Diabetes 6(1):37–53 Lei W, Peder CP, Diane MS, Bengisu T, Emmanuel A, Ron I, Qian H (2015) An automated assessment system of diabetic foot ulcers based on wound area determination, colour segmentation, and healing score evaluation. J Diabetes Sci Technol 10(2):421–428 Mathers CD, Loncar D (2006) Projections of global mortality and burden of disease from 2002 to 2030. PLoS Med 3(11):e442 Pavicic T, Korting HC (2006) Xerosis and callus formation as a key to the diabetic foot syndrome: dermatologic view of the problem and its management. J Dtsch Dermatol Ges 4(11):935–941 Prabhu KG, Patil KM, Srinivasan S (2001) Diabetic feet at risk: a new method of analysis of walking foot pressure images at different levels of neuropathy for early detection of plantar ulcers. Med Biol Eng Comput 39(3):288–93
Heavy Metal Ions Detection by Carbon Paste Electrode as an Electrochemical Sensor Arti Mourya, Bidyut Mazumdar, and Sudip K. Sinha
Abstract Voltammetry techniques such as cyclic voltammetry and electrochemical impedance spectroscopy are used to quantify metal ions in aqueous solutions. In this paper, we have reported electrochemical detection of lead (II), cadmium (II), and Copper (II) ions based on carbon paste electrode using graphite powder. Potassium ferricyanideK3 [Fe(CN)6] solutions as benchmark media to check redox reactions. The Detection limit was 6.56 × 10–9 M investigated by cyclic voltammetry. Effects of pH and supporting electrolytes were also measured. Keywords Carbon paste electrode · Cyclic voltammetry · Electrochemical impedance spectroscopy · Heavy metal
1 Introduction Water contamination is one of the major environmental issues and between the wide diversity of contaminants heavy metal ions are one of them Momodu (2010) Heavy metal contamination damages the ecosystem due to higher toxicity even at minute quantity. Cadmium (II), Lead (II), and Copper (II) are classified as heavy metal ions which enter in water bodies through various industrial operations. Elevated levels of heavy metal ions can cause behavioral changes, liver damage, and impair intelligence (Migliorini 2017; Rajawat 2014). Hence, the recognition and monitoring of toxic metals from polluted sites and aquatic ecosystems is an important analytical task that is needed by society. Various analytical methods and techniques are used to determine the heavy metal ions at low concentrations like atomic absorption/emission spectrometry
A. Mourya · B. Mazumdar (B) Department of Chemical Engineering, NIT Raipur, Raipur, India e-mail: [email protected] S. K. Sinha Department of Metallurgy Engineering, NIT Raipur, Raipur, India © Springer Nature Singapore Pte Ltd. 2021 A. A. Rizvanov et al. (eds.), Advances in Biomedical Engineering and Technology, Lecture Notes in Bioengineering, https://doi.org/10.1007/978-981-15-6329-4_4
29
30
A. Mourya et al.
(AAS/AES), and atomic fluorescence spectrometry (AFS). Among them electroanalytical method is superior, because of its high sensitivity, eco-friendly, selectivity, easy data readout, quick response time, cost-effective, and easy handling (Vasileva 2017; Rajesh 2017). Selecting suitable material for electrochemical senor decides good and reliable detection for heavy metal ions (Bagheri 2013). Graphite is a unique and versatile material in electrochemical systems, has been utilized for decades in sensors, batteries, fuel cells, and electrochemical capacitor Panice (2014). In this reporting, we study on simulated aqueous phase (with lead, cadmium, and copper) by carbon paste electrode (CPE) incorporated with voltammetric methods.
2 Material and Methods 2.1 Reagents All the experiments are carried out with the help of analytical grade chemicals without any further purification. Graphite powder, chemicals (size < 40 µm), paraffin liquid, and aluminum powder 98% were purchased from Loba chemicals. K3 Fe(CN)6 was purchased from Sigma-Aldrich. Cadmium nitrate, lead nitrate, cupric nitrate, potassium nitrate 98% (KNO3 ), and NaOH chemicals are obtained from Merck.
2.2 Preparation of Electrode as a Sensor For the preparation of CPE, 70% graphite and 30% paraffin oil binder was mixed in mortar and pestle. The paste was left overnight for self homogenization. After that, the resulting paste was filled into an insulin syringe (electrode holder). A thin copper wire was inserted through the opposite end of the electrode body for the electrical contact. And the surface of the electrode gets smoothed on A4 paper until it had a mirror finish. After each experiment, a few millimeters (mm) of carbon paste was extruded from the electrode holder. Further it was polished by 0.05 micron alumina powder slurry, and washed well by distilled water and ethanol (Randelovi´ c 2017).
3 Result and Discussion 3.1 Cyclic Voltammetry To check the electrochemical properties of CPE, cyclic voltammetry (CV) technique was used for 5 × 10 − 3 mol L − 1 K3 Fe(CN)6 and 1 mol L − 1 KNO3 as an electrolyte. Figure 1a shows the cyclic voltammogram at different scan rates with
Heavy Metal Ions Detection by Carbon Paste Electrode …
31
Fig. 1 a Cyclic voltammogram of CPE for 5 × 10−3 mol L−1 K3 Fe(CN)6 and 1 mol L−1 KNO3 at different scan rates. b potential versus current at different scan rates (0.1–0.5 mV/s), c calibration curve at different concentration of K3 Fe(CN)6
single anodic and cathodic peak currents. A plot between potential versus current Fig. 1b for different scan rates shows a linear equation with 0.95 and 0.97 regression values. Randles–Sevcik equation was used to calculate the electrode surface area: Ip = (2.69 × 105)n3 /2AD1/2V1/2, where, A surface area of the working electrode (cm2 ), Ip peak current (A), D diffusion coefficient, (cm2 /s), C initial concentration (mol/L), n number of electrons involved in the reaction, and V scan rate (V/s). Further, Fig. 2a, b shows cyclic voltammograms for the simultaneous determination of the lead, cadmium, and copper ions at a different scan rates (10–50 mV/s) and concentrations (1–7 mM) in pH of 5.23. To investigate the reaction kinetics, the effect of scan rate on the electrochemical redox behavior, CV shows three different anodic and cathodic peaks for lead (II), cadmium (II), and copper (II) ions. It can be clearly seen that the anodic and cathodic peak current increased with increasing the concentrations of heavy metal ions.
32
A. Mourya et al.
Fig. 2 Cyclic voltammetry of carbon paste electrode at a different scan rates (0.1–0.5), b different concentration (1-7 mM).
Fig. 3 a Effect of pH and b electrolytes on cyclic voltammogram
3.2 Effect of Operating Parameters Effect of pH &Electrolyte. The voltammetric output was investigated between the pH range of 2.0 -6.0 in 1 M KNO3 electrolytic solution with 5 mM of Pb, Cd, and Cu. As expected (Fig. 3a), higher current signals were observed at lower pH. However, the peak current was decreased because of the hydrolysis of heavy metal ions as the pH reaches higher value from 2.0 to 6.0 (Xiong 2016). The electrochemical behaviors of the heavy metal ions are different in different electrolytes such as HCl, KCL, KNO3 , NaNO3 , and acetic acid. Among these electrolytes (Fig. 3b), HCL shows the higher current signals due to the occurrence of well-defined peaks with the largest cathodic peak current of 4 mA.
Heavy Metal Ions Detection by Carbon Paste Electrode …
33
Fig. 4 Electrochemical impedance spectroscopy of carbon paste electrode containing 5.0 mM Pb, Cd, and Cu solution with 1 M KNO3
Table 1 Impedance parameter by CPE for Pb(II), Cd (II), and Cu (II) Parameters
Q-Yo
Q-n
R
Q-Yo
Chi squared
Electric circuit
CPE
3.252E-8
0.8
5590
0.0001203
2.370e-02
(Q(RQ)
3.3 Electrochemical Impedance Spectroscopy (EIS) The EIS was performed after a (70 s) open circuit potential, and 0.1–100000 Hz frequency ranges. The Nyquist plot (Fig. 4) represents the electrical equivalent circuit (inset) used in the EIS fitting and impedance parameters of the CPE are shown in Table1. From the results obtained, CPE showed a semicircle in lower frequency and a straight line with higher range. According to open literature such behavior shows a charge transfer limited process and its value is equal to the diameter of the semicircle.
4 Conclusion CPE was effectively developed for electrochemical detection of lead (II), cadmium (II), and copper (II) ions.The CV results show successive enhancement in the peak currents (Ip) with a constant potential. As expected, higher current signals were observed at lower pH value of 2. (Q(RQ) electrical circuit was a best fit model for electrochemical impedance spectroscopy, performed to check the interface properties of the prepared electrode. Finally the developed sensor was successfully applied for heavy metal ion detection in aqueous solution by cyclic voltammetry technique.
34
A. Mourya et al.
References Bagheri H (2013) Simultaneous electrochemical determination of heavy metals using a triphenylphosphine/MWCNTs composite carbon ionic liquid electrode. Sensors Actuators B Chem 186:451–546 Migliorini F (2017) Voltammetric cadmium (II) sensor based on a fluorine doped tin oxide electrode modified with polyamide 6/chitosan electrospunnanofibers and gold nanoparticles. Microchim Acta 184(4):1077–1084 Momodu MA (2010) Heavy metal contamination of ground water: the case study. Res J Environ Earth Sci 2(1):39–43 Panice LB (2014) Electrochemical properties of the hexacyanoferrate (II)–ruthenium (III) complex immobilized on silica gel surface chemically modified with zirconium (IV) oxide. Mater Sci Eng B 188:78–83 Penka V (2017) Application of starch-stabilized silver nanoparticles as a colorimetric sensor for mercury(II) in 0.005 mol/L Nitric Acid. J Chem 1–10 Rajawat DS (2014) Trace determination of cadmium in water using anodic stripping voltammetry at a carbon paste electrode modified with coconut shell powder. J Anal Sci Technol 5(1):19 Rajesh K (2017) Design and development of graphene intercalated V2 O5 nanosheets based electrochemical sensors for effective determination of potentially hazardous 3, 5–Dichlorophenol. Mater Chem Phys 199:497–507 Randelovi´ c MS (2017) Electrocatalitic behaviour of serpentinite modified carbon paste electrode. J Electroanal Chem 801:338–344 Shrivastava A (2011) Methods for the determination of limit of detection and limit of quantitation of the analytical methods. Chronicles Young Sci 2(1):21 Xiong W (2016) Development of gold-doped carbon foams as a sensitive electrochemical sensor for simultaneous determination of Pb (II) and Cu (II). Chem Eng J 284:650–656
Median Filtering Detection Using Markov Process in Digital Images Saurabh Agarwal and Satish Chand
Abstract Digital images are prevalent medium for information representation. The wide acceptability of images also leads to some problems. Few people spread forged images for their own interests. This type of practice questions the credibility of images. To ensure the credibility, forensic investigation is carried out to ensure the genuineness of images. Detection of median filtering is one of the important forensic analyses. Median filtering has several uses in antiforensics. It is applied to hide the impression of image forgery due to its nonlinear nature. Some techniques are evolved to countering these antiforensics i.e., median filtering. These techniques can classify median filter operated and non-median filter operated images. In this paper, we compare three techniques that are based on Markov process. We find that performance of these techniques varies according to the size of database. Keywords Image forensics · Median filtering detection · Antiforensic
1 Introduction From earlier time, pictures are the preferable medium of information representation. In this digital era, most of the modern gadgets are equipped with camera. In spite of this, with the help of high-speed internet, images can be communicated anywhere. Sometimes peoples float forge images for their own malicious interests. We can see the latest example of image forgery popular on twitter November 1, 2017 in Fig. 1. Figure 1a is a pristine image from the scene of movie “Kya Yeh Sach Hai” and Fig. 1b is a forged image. In this forged image, it is shown that a senior police officer kneeling and touching the feet of Home Minister Rajnath Singh. However, it is found
S. Agarwal (B) ASET, Amity University, Noida, India e-mail: [email protected] S. Chand Jawaharlal Nehru University, Delhi, India © Springer Nature Singapore Pte Ltd. 2021 A. A. Rizvanov et al. (eds.), Advances in Biomedical Engineering and Technology, Lecture Notes in Bioengineering, https://doi.org/10.1007/978-981-15-6329-4_5
35
36
S. Agarwal and S. Chand
Fig. 1 An example of pristine and fake image
(a) Pristine Image
(b) Fake Image
that some opponent morphed Rajnath Singh’s face onto the face of the actor playing the minister role in the scene of movie “Kya Yeh Sach Hai.” To ensure the credibility of images, forensic analysis of images is performed. For creating the realistic forge image, various post-processing operations are performed on forged image like resampling, contrast enhancement, histogram equalization, etc. Some of these post-processing operations have also been used to hide the forgery artifacts and this process is called antiforensics. The forensic analysis of various antiforensics operations is necessary to identify image forgery. Out of these operations, detection of median filtering is crucial and difficult due to its nonlinear nature. The importance of median filtering motivates researchers to develop several methods for detecting it. The first well known method (Kirchner and Fridrich 2010) detects the median filtering by considering streaking artifacts in histogram bins of first-order difference images. This method fails to provide good accuracy on compressed images. The method (Pevny 2010) is known as Subtractive Pixel Adjacency Matrix (SPAM) applicable for many applications like Steganalysis is also
Median Filtering Detection Using Markov Process …
37
applied for median filtering detection. It gives good accuracy on compressed images also. In this method Markov process is applied on first-order difference images. Markov process provides Subtractive Pixel Adjacency Matrix (SPAM) features on adjacent difference pairs. The method (Chen 2013) known as Global and Local Feature set (GLF) extracts local correlation of neighboring pairs on first- and secondorder difference image pixels. The authors have shown some results on noisy images and claim better results from previously available techniques. The technique (Shen 2014) utilizes Rotation-Invariant Local Binary Pattern (RI-LBP) texture descriptor for median filtering detection. Similarly, the technique (Zhang 2014) uses higher order Local Ternary Pattern (LTP) texture descriptor to identify median filtering. These techniques (Shen 2014; Zhang 2014) provide average results on low resolution and compressed images. In (Ravi et al. (2015)) transition probability features are extracted from image by utilizing the correlation of spatial domain quantization noise. This method provides good results on non-compressed and low compressed images. Further method SPHO (Agarwal 2016) based on SPAM considered higher order pixel differences for improving the detection accuracy. The method (Agarwal 2016) provides better results on low resolution and highly compressed images. In this paper, we assess the performance of methods (Agarwal 2016); Chen 2013; Pevny 2010) on different sizes of image databases. This paper primarily emphasizes on the effect of image database size on median filtering detection accuracy. We have chosen these methods (Agarwal 2016; Chen 2013; Pevny 2010) because their performance is somehow comparable.
2 Feature Construction To differentiate between median filtered and non-filtered images, relevant features need to be extracted from images. In SPAM Pevny (2010), features are extracted using Markov process on first-order difference arrays. In SPHO Agarwal (2016) Markov process is applied on first, second, and third order difference arrays. The technique GLF TECH (Chen 2013) extracts two types of feature sets i.e., global probability feature set and local correlation feature set. In which global probability feature set extraction process. It can be seen that somehow these three methods are based on Markov process. Therefore, here we discuss briefly about Markov process when applied on difference arrays. First-order backward difference array in horizontal direction is defined for a 2–D image X of size m × n as: h = X u,v − X u+1,v Du,v
(1)
Where, u ∈ {1, 2, . . . , m − 1}, v ∈ {1, 2, . . . , n}. Like first-order difference array in horizontal direction, difference array in forward direction can be found. Similarly, forward and backward differences for major diagonal, minor diagonal, and vertical directions can be identified.
38
S. Agarwal and S. Chand
Now, Markov process of second order is applied on these difference arrays. The Markov process of second order is used in further discussions and experiment works. Markov process of second order provides relevant size and high-quality discriminant features. The second-order Markov process is defined as. M p,q,r = pr (Du+2,v = P|Du + 1, v = q.Du,v = r )
(2)
where Pr represents probability p, q, r ∈ {−T, . . . , T } and for Pr(Du+1,v = q, Du,v = r ) = 0, M p,q,r = 0. Here T is the threshold parameter. In thresholding process, all values greater than T is replaced by T and all values smaller than −T is replaced by −T in difference array D. Therefore, we have numbers in the range of {−T, −T + 1, …, 0,…, T−1, T}. This helps in obtaining small size feature array. The feature array is found by concatenation of two feature arrays. The first feature array contains the Markov features of horizontal and vertical difference 2–D arrays and the second feature array contains Markov features of major diagonal & minor diagonal difference 2–D arrays, in forward and backward directions. Finally, the second-order Markov process provides the feature array of size 2 * (2T + 1)3.
3 Experimental Setup and Results In this section, we will discuss image databases, classifier, and experimental results.
3.1 Datasets We evaluate the performance of SPAM Pevny (2010), SPHO (Agarwal 2016), and GLF TECH (Chen 2013) on combined databases of BOWS2 Bas and Furon (2008), UCID (Schaefer and Stich 2003), and NCI (Liu and Chen 2014). The BOWS2 grayscale image database contains 10,000 images of size 512 × 512 pixels. The UCID color image database consist 1,338 images of size 384 × 512 pixels. The NCI color image database has 5,150 images of size 256 × 256 pixels. These databases have images of several types like indoor, outdoor, natural, objects, fruits, foods, textures, etc. We crop the images by taking central block of size 64 × 64 pixels from all three used databases (Bas and Furon 2008; Liu and Chen 2014; Schaefer and Stich 2003). From these images, two sets of images are produced. The first set is created by applying JPEG compression of quality 50. The second set is created by applying median filter of size 3 × 3 and JPEG compression of quality 50. These sets are named as set 1 and set 2, respectively. From these two sets, we select equal number of images for performing experiments.
Median Filtering Detection Using Markov Process …
39
3.2 Classifier We have applied Linear Discriminant Analysis (LDA) classifier to distinguish between median and non-filtered images. LDA classifier and Markov process both consider the conditional joint distribution. Therefore, LDA classifier is more appropriate choice than Support Vector Machine (SVM) classifier. In experiments, 50% images are considered for training and rest 50% for testing from the set of images of both types i.e., median and non-filtered images set. We have shown the experimental results in terms of detection accuracy (AC) and ROC curve with classification error (Pe). Detection accuracy (AC) in percentage (%) is described as: AC =
(TP + TN) ∗ 100 (TP + TN + FP + FN)
(3)
where, TP-True Positive, TN-True Negative, FP-False Positive, and FN-False Negative. Classification error (Pe) = 1 − (AC/100).
3.3 Experimental Results First, we have shown the results by taking 500 non-filtered images from set 1 and 500 median filtered images from set 2, randomly. As can be seen from Fig. 2a SPAM method for T = 2 provides the least classification error i.e., 9.44. Only GLF TECH method is comparable with SPAM with 11.21 classification error. Method SPHO for T = 3 performs worst with 31.52 classification error (Pe). In the second experiment, we have taken 1000 images from each set. It is clearly seen from Fig. 2b, method SPHO for T = 2 and SPAM for T = 2 perform decently with Pe = 5.45 & 6.32, respectively. Similarly, we evaluate the performance by considering 2000, 3000, 4000, & 5000 from both sets i.e., set 1 & set 2. As evident from Figs. 2c–f method SPHO T = 2 performs best. As can be seen from Fig. 3 SPHO for T = 2 gives best results for more than 2000 images consistently. The performance of SPAM for T = 2, 3 is also fair. GLF TECH performance is consistent but detection accuracy is least in most of the cases. The performance of SPHO for T = 3 is worst for small number of images.
4 Conclusion High dependency and usage of digital images make it more vulnerable. Strong antiforensics techniques are required to compete with forgers. One of the popular antiforensics operations is median filtering. In many cases, detection of median filtering is crucial for forensic investigation of images. A number of methods have
40
S. Agarwal and S. Chand
Fig. 2 Median filtering detection techniques ROC curves with classification error
(a) Performance comparison for 500 images of set 1 and set 2
(b) Performance comparison for 1000 images of set 1 and set 2
(c) Performance comparison for 2000 images of set 1 and set 2
Median Filtering Detection Using Markov Process …
41
Fig. 2 (continued)
(d) Performance comparison for 300 images of set 1 and set 2
(e) Performance comparison for 4000 images of set 1 and set 2
(f) Performance comparison for 5000 images of set 1 and set 2
42
S. Agarwal and S. Chand
Median Filtering DetecƟon
DetecƟon Accuracy (%)
100 95 90 85
SPAM T=2
80
SPAM T=3
75
SPHO T=2
70
SPHO T=3
65
GLF TECH
60
1000
2000
4000 6000 8000 Number of Images
10000
Fig. 3 Comparative analysis of methods with respect to number of images
been discussed for median filtering detection. In this paper, the performance of three popular methods is compared on diverse databases. We find that the size of database affects the performance of these methods. In most of the cases, classification model build from large size of database provides better detection accuracy. Acknowledgements The authors are thankful to creator of BOWS2 [2], UCID [8] and NCI [5] for sharing their databases. They are also thankful to authors of SPAM [6], and GLF TECH [3] for providing their code.
References Agarwal S et al (2016) SPAM revisited for median filtering detection using higher-order difference. Secur Commun Netw 9(17):4089–4102 Bas P, Furon T (2008) Break our watermarking system. Available https://bows2.ec-lille.fr/2nd Chen C et al (2013) Blind detection of median filtering in digital images: a difference domain based approach. IEEE Trans Image Process 22(12):4699–4710 Kirchner M, Fridrich J (2010) On detection of median filtering in digital images. In: Media forensics and security II, p 754110 Liu Q, Chen Z (2014) Improved approaches with calibrated neighboring joint density to steganalysis and seam-carved forgery detection in JPEG images. TIST 5:4 Pevny T et al (2010) Steganalysis by subtractive pixel adjacency matrix. IEEE Trans Inf Forensics Secur 5(2):215–224 Ravi H et al (2015) Spatial domain quantization noise based image filtering detection. In: 2015 IEEE international conference on image processing (ICIP), IEEE, pp 1180–1184 Schaefer G, Stich M (2003) UCID—an uncompressed colour image database. SPIE, Storage Retr Methods Appl Multimed 5307:472–480 Shen Z et al (2014) Blind detection of median filtering using linear and nonlinear descriptors. Multimed Tools Appl 75(4):2327–2346 Zhang Y et al (2014) Revealing the traces of median filtering using high-order local ternary patterns. IEEE Signal Process Lett 21(3):275–280
Differential of EMG Activity of Selected Calf Muscle During DLHR Exercise in Relation to Performance Level Monika, L. M. Saini, and Saravjeet Singh
Abstract Muscular fatigue is described as a condition when the ability of muscles to contract and produce force is reduced under sustained contraction. The quantification of muscular fatigue by surface electromyography (EMG) provides a noninvasive method for easily accessing and measuring the physiological processes occurring during sustained muscular work. Double leg heel raise (DLHR) exercise is performed for the strengthening of calf muscles which generally gets weak following immobilization after injury or surgery. It is used as evaluation process/test by physiotherapists/clinicians/others associated with rehabilitation of athletes/nonathletes to strengthen their lower-body muscles and connective tissues after joint-related injury. The study purpose is to compare the effect of sustained DLHR exercise in relation to performance level among males. The EMG activity of Gastrocnemius Lateral (GSL) and Gastrocnemius Medial (GSM) muscles of both legs are considered for this study because they are the dominating calf muscle. Here frequency domain and time-domain features were chosen for extracting necessary information from the EMG signal using the algorithm developed in MATLAB. Understanding from the findings is significant in order to achieve optimum muscular strength and strength endurance development whereas reducing the probability of training-related injuries to sportspersons. Keywords Gastrocnemius lateral (GSL) · Double leg heel raise · Right leg (RL) · Fatigue · Mean frequency
Monika (B) · L. M. Saini School of Biomedical Engineering, NIT Kurukshetra, Kurukshetra, Haryana, India e-mail: [email protected] S. Singh Department of Biomedical Engineering, DCRUST Murthal, Murthal, Haryana, India © Springer Nature Singapore Pte Ltd. 2021 A. A. Rizvanov et al. (eds.), Advances in Biomedical Engineering and Technology, Lecture Notes in Bioengineering, https://doi.org/10.1007/978-981-15-6329-4_6
43
44
Monika et al.
1 Introduction In the past decade, fatigue is explained as an exercise-induced, decrease in the muscle ability to produce power/force, if or not the task can be continued (Barry and Enoka 2007). When we exercise intensely/carry on similar kind of motion activity, then muscle of our body fatigue and as a result, we become to lose coordination, cannot train efficiently, work efficiency potential decreases for serious injury (Sakurai et al. 2010). So, muscle fatigue is essential information during muscular strength building or working, and is necessary to evaluate quantitatively. The sEMG i.e. surface electromyography (electrical activities measured over the skin surface during muscle contractions) signal analysis helps in prosthetic myoelectric control, clinical diagnosis, ergonomics, sports biomechanics, and evaluating the muscle fatigue (Karthick et al. 2018). The Gastrocnemius muscle (overlying the soleus) is a broader calf muscle which makes half of the calf muscle. It runs down back of leg, from behind knee to the Achilles tendon in the heel. The gastrocnemius muscle consists of 2 heads: lateral and medial, join soleus at tendo-Achilles and inserts distally on the calcaneus. Gastrocnemius crosses 2 joints, working for plantar flexion of the ankle and flexion of knee. Muscles cross two joints, like gastrocnemius muscle, are especially susceptible to injury. Gastrocnemius injuries may happen at distal muscle–tendon junction at the head of medial muscle. The medial head of gastrocnemius may fracture during jumping or sprinting activities. Affected athletes suddenly have stabbing pain and may feel like their back of the calf has been struck directly with a ball. Shield Against Calf Sprains—According to a survey conducted by Runner’s World (in 2011), calf sprains were the 2nd most common injury between injured runners. When calf muscles are not trained regularly, it loses strength and conditioning which is necessary for it in order to support athletic activities and as a result chance of injury increases. During running, walking, or jumping, our calf muscles, mainly the gastrocnemius muscle generates power (California 2005). In sports like volleyball and basketball, stronger calf muscles may be advantageous for the player. Strengthening of calf muscle is necessary in order to prevent any injury from accidentally turning of foot outwards or inwards. Double leg heel raise (DLHR) exercise is performed for the strengthening of calf muscles (California 2005) which generally gets weak following immobilization after injury or surgery. Injuries, like Ankle fractures and Achilles tendon rupture, usually require a concerted calf stretching effort during recovery period. This exercise also strengthens the tibialis posterior muscle, which gets weak/dysfunctional in patients having acquired adult flatfoot deformity. The study purpose is to compare the effect of sustained DLHR exercise in relation to performance level among males. The EMG activity of Gastrocnemius Lateral (GSL) and Gastrocnemius Medial (GSM) muscles of both legs are considered for this study because they are the dominating calf muscle. Here frequency domain and timedomain features were chosen for extracting necessary information from the EMG signal using the algorithm developed in MATLAB. We used EMG signal because it
Differential of EMG Activity of Selected Calf …
45
reflects the degree of muscular activation (Hopkins et al. 1999). Also, EMG signal is widely used in the human movement analysis where myoelectric signals produced by muscles are analyzed for muscular function investigation (Rainoldi et al. 2004). Understanding from the findings is significant in order to achieve optimum muscular strength and strength endurance development whereas reducing the probability of training-related injuries to sportspersons (Kaur et al. 2016).
2 Material and Methodology Five healthy sports people engaged in different sports and six normal people were recruited as volunteers (age = 21.5 ± 1.87, weight = 59 ± 10.21 kg, height = 172.96 ± 8.55 cm). Criteria of Inclusion for the study were subjects above 18 years of age, nonalcoholic, willing to engage in the study, and not having any other medical diseases which might hamper their nerves and muscles functioning, in order to make them unfit for participation in the study. Before their participation in the study, each member was described about the motive and protocol to be followed. In order to make them familiar with the study, the participants were asked to experience few pre-data collection trails. As per ethical guidance and needs, all participants provided voluntary written informed consent before their participation in the study.
2.1 Data Acquisition EMG data were obtained using a 4-channel EMG BIOPAC Inc. MP 100 system (Gain: 5–50,000, CMRR: 110 dB at 50/60 Hz and Input Impedance: 2 M) in a quiet laboratory room for all the subjects. Before data acquisition, skin of subjects was first shaved to remove hairs and then cleaned with cotton carrying alcohol for minimizing skin impedance and enhancing signal acquisition. Then, Disposable electrodes (7.5 mm diameter) were placed over the Gastrocnemius Lateral (GSL) and Gastrocnemius Medial (GSM) muscles of both legs of the participants based on Seniam (European Recommendations for Surface Electromyography) (Disselhorst-klug 1999). The EMG data were obtained from both the legs simultaneously. Reference electrode was put on the ankle of the subject to act as ground and for safety demands of the equipment (Konrad 2006). Inter-electrode separation (center to center) was kept 20 mm. The schematic of electrode placement is shown in Fig. 1. After the subject preparation, they were instructed to perform Double Leg Heel Raise (DLHR) exercise for 120 s in order to become familiar with the protocol. The sampling frequency of data acquisition software (Acqknowledge3.9.9, BIOPAC Systems Inc.) was set to 2000 Hz in order to avoid the aliasing effects as per Nyquist criteria. Data Segmentation. The study group of male athletes and nonathlete subjects was divided into high and low performance groups, respectively. These groups were given
46
Monika et al.
Fig. 1 Electrode placement and muscles
name as Male high performance (MHP) and Male low performance (MLP). All these groups were able to perform a minimum of 120 s DLHR exercise, hence for better understanding the influence of performance on EMG signal, data is further divided into 40, 80, and 120 s. Data processing. Acquired raw EMG signals from different subjects were quantified into MATLAB. The signal processing methodologies applied to the raw EMG signals include two parts namely preprocessing and feature extraction as depicted in Fig. 2. The Digital filters were used in MATLAB for EMG signal preprocessing (Kaur et al. 2016). For the removal of 50 Hz noise interference from EMG signal, Notch filter was applied. To remove other noise sources from EMG signal, cascaded high pass filter (20 Hz) and low pass filter (450 Hz) were applied (Luca et al. 2010). Time-domain and Frequency-domain features were extracted from the filtered EMG signal namely Root Mean Square (RMS), Integrated EMG (IEMG), Median Frequency (MDF), and Mean Frequency (MNF), respectively for each subject. The reason for choosing these features in this study is that these are the most often selected features for the determination of changes due to fatigue in muscles. Integrated EMG (IEMG). It represents the sum of the absolute value of the EMG signal amplitude (Phinyomark et al. 2012). IEMG relates to the EMG signal firing point and mathematically represented as: Xi (t) =
XR (t)dt
(1)
where Xi (t) is integration of signal and XR (t) is the rectified signal. Increase of IEMG value during contraction of muscle corresponds to higher muscle fiber recruitment for a fixed external force. Therefore, its analysis is necessary in order to determine the fatigue development in muscle fiber(s) during contraction.
Differential of EMG Activity of Selected Calf …
47
Fig. 2 EMG signal processing steps
Root Mean Square (RMS). This feature is most advanced and is used to analyze EMG signal. RMS is framework as amplitude modulated Gaussian random process which corresponds to non-fatiguing contractions and constant force (Phinyomark et al. 2016). RMS is the square root of mean power of EMG signal for a given time period and reflects the signal mean power. This feature is considered to be analyzed here because it shows the extent of physiological activity in motor unit during sustained contraction of muscle (Konrad 2006). Mathematically it is computed as: T2 1 XR (t)2 (2) Xr (t) = T2 − T1 T1 Mean Frequency (MNF). Mean Frequency is defined as the summation of product of EMG power spectrum and the frequency divided by the total sum of spectrum intensity (Phinyomark et al. 2010) and can be written as: M
(fm Pm )/
m=1
M
(Pm )
(3)
m=1
Here f m is the spectrum frequency at frequency bin m, Pm describes the EMG power spectrum at frequency bin m, and M is the frequency bin length. This feature is used in the estimation of muscle fatigue in EMG signal and hence is analyzed in
48
Monika et al.
this study. MNF of EMG signal is double when muscle is resting in comparison to EMG signal of muscle under fatigue (Thongpanja et al. 2013). Median Frequency (MDF). MDF is the frequency where the spectrum splits into two equal amplitude regions (Phinyomark et al. 2010) and is half of Total Power (TTP) Feature (Phinyomark et al. 2012). It is calculated in two levels as: First, the signal intensity in whole spectrum is summed and then divided by two. In the second level, a frequency is selected where cumulative intensity (i.e., all intensity values of all frequencies lower than the selected frequency including focal intensity) exceeds the value calculated in level 1. Muscle fatigue causes a downward shift of EMG signal frequency spectrum (Pincivero et al. 2018). It is the most applicable feature and often used for describing fatigue behavior of muscle. MDF is calculated as: MDF
Pm =
m=1
M m=MDF
Pm =
M 1 Pm 2 m=1
(4)
Here Pm describes the EMG power spectrum at frequency bin m, and M describes the frequency bin length.
3 Result and Discussion IEMG and RMS values of GSL and GSM muscles of Left Leg (LL) and Right Leg (RL) are given below. It demonstrates performance difference for male athletes and nonathletes with respect to time-domain features of both the muscles for both legs. Higher performance groups were having higher mean value for IEMG and RMS in terms of performance difference along with time progression in both muscles of subjects during DLHR exercise (Fig. 3). Mean Frequency and Median Frequency values of GSL and GSM muscles of Left Leg (LL) and Right Leg (RL) are shown below. It presents performance difference for male athletes and nonathletes in the experimental protocol with respect to the GSL and GSM muscles activity for both legs. Lower performance groups were having higher mean value for MNF and MDF (except MDF of GSMof Right Leg (RL) where Higher performance groups were having higher mean value) in terms of performance difference along with time progression in both muscles of subjects during DLHR exercise (Fig. 4).
Differential of EMG Activity of Selected Calf …
49
RMS of GSM (LL) at different 0.09 time
0.15 0.1 0.05 0
MLP
MLP
0.075
MHP
0.07
MHP
40 80 120 Time (seconds)
RMS of GSL (RL) at different time 0.085 0.08 0.075 MLP 0.07 MHP 0.065 40 80 120 Time (seconds)
RMS of GSM (RL) at different 0.2 times 0.15 0.1 MLP 0.05 MHP 0 40 80 120 Time (seconds)
IEMG of GSM (LL) at different time 4200 4000 3800 MLP 3600 MHP 3400 3200 40 80 120 Time(seconds) IEMG of GSM (RL) at different time 10000
MLP
3400
MHP
3200 40 80 120 Time (second)
Mean
3600
8000 6000 4000 2000 0
MLP MHP 40 80 120
IEMG of GSL (LL) at different time 5000 4000 MLP 3000 2000 MHP 1000 0 40 80 120 Time (seconds)
Mean
Time (seconds)
IEMG of GSL(RL) at different time 3800 Mean
0.08
Mean
Mean
Mean
40 80 120
0.085
Mean
Mean
RMS of GSL(LL) at different time
Time (seconds)
Fig. 3 Comparison of Time-domain features (RMS and IEMG) in terms of performance level
Monika et al.
MNF of GSL (RL) at different me 220 210 200 MLP 190 MHP 180
MNF of GSM (LL) at different me 210 200 190 MLP 180 MHP 170 160 40 80 120
Mean
MNF of GSL (LL) at different me 260 250 240 MLP 230 MHP 220 210 200 40 80 120 Time (seconds)
Time (seconds) MNF of GSM (RL) at different mes 230 220 MLP 210 MHP 200 190 40 80 120 Time (seconds)
Mean
Mean
Mean
50
170 40 80 120
Mean
MDF of GSL(RL) at different me 250 200 MLP 150 100 MHP 50 0 40 80 120 Time(seconds)
MDF of GSM (LL) at different me 200 190 180 MLP 170 MHP 160 150 140 40 80 120 Time (seconds) MDF of GSM(RL) at different me 200
Mean
Mean
MDF of GSL (LL) at different me 300 250 200 MLP 150 100 MHP 50 0 40 80 120 Time (seconds)
Mean
Time (seconds)
190
MLP
180
MHP
170 160 40 80 120 Time(seconds)
Fig. 4 Comparison of Frequency-domain features (MNF and MDF) in terms of performance level
Differential of EMG Activity of Selected Calf …
51
4 Conclusion Difference in EMG activity of left and right leg, muscle to muscle as well as Time-domain and Frequency-domain features extracted were observed. The Performance level of male athletes and nonathletes exhibits insignificant EMG activity differences of selected muscles. At last, we conclude that understanding from the findings is significant in order to achieve optimum muscular strength and strength endurance development whereas reducing the probability of training-related injuries to sportspersons.
References Barry BK, Enoka RM (2007) The neurobiology of muscle fatigue: 15 years later. Integr Comp Biol 47(4):465–473 California S (2005) Biomechanics of the heel-raise exercise biomechanics, Apr 2014 De Luca CJ, Gilmore LD, Kuznetsov M, Roy SH (2010) Filtering the surface EMG signal: movement artifact and baseline noise contamination. J Biomech 43(8):1573–1579 Disselhorst-klug C (1999) European recommendations for surface ElectroMyoGraphy, pp 8–11 Hopkins JT, Ingersoll CD, Sandrey MA, Bleggi SD (1999) An electromyographic comparison of 4 closed chain exercises 34(4):353–357 Karthick PA, Ghosh DM, Ramakrishnan S (2018) Surface electromyography based muscle fatigue detection using high-resolution time-frequency methods and machine learning algorithms. Comput Methods Programs Biomed 154:45–56 Kaur M, Nara S, Shaw D, Bhatia D (2016) EMG asymetricity of selected knee extensor muscles in sustained Squat posture (a Yogic posture) of athletes in relation to their gender and performance 7:1–6 Konrad P (2006) The ABC of EMG, Mar 2006 Phinyomark A, Hirunviriya S, Limsakul C, Phukpattaranont P (2010) Evaluation of EMG feature extraction for hand movement recognition based on Euclidean distance and standard deviation. In: 2010 international conference on electrical engineering, computer, telecommunications and information technology (ECTI-CON), pp 856–860 Phinyomark A, Phukpattaranont P, Limsakul C (2012) Feature reduction and selection for EMG signal classification. Expert Syst Appl 39(8):7420–7431 Phinyomark A, Limsakul C, Phukpattaranont P (2009) A novel feature extraction for robust EMG pattern recognition 1:71–80 Pincivero DM et al (2018) Influence of contraction intensity, muscle, and gender on median frequency of the quadriceps femoris 99004:804–810 Rainoldi A, Melchiorri G, Caruso I (2004) A method for positioning electrodes during surface EMG recordings in lower limb muscles 134:37–43 Sakurai T, Toda M, Sakurazawa S, Akita J, Kondo K, Nakamura Y (2010) Detection of muscle fatigue by the surface electromyogram and its application 43–47 Thongpanja S, Phinyomark A, Phukpattaranont P, Limsakul C (2013) Mean and median frequency of EMG signal to determine muscle force based on time-dependent power spectrum 51–56
Advanced Encryption Standard Algorithm in Multimodal Biometric Image Sharmila S. More, Bhawna Narain, and B. T. Jadhav
Abstract Cryptography is becoming slowly but surely a more important feature of computer security. In every field of technology, there are major issues of security but these issues overcome by using cryptographic algorithms. Biometric images are used as input and then the features like shape, size, edge, and texture are extracted. This is done by applying feature extraction algorithms and advanced encryption standard algorithm on Fingerprints, Iris, faces, and palm print simultaneously. In this paper, we have explained the information about multimodal biometrics, limitations of the DES algorithm on the encryption process of multimodal biometric images, Solution to these problems, and the role of AES in multimodal. We have discussed the implementation of the AES algorithm by using MATLAB on parameters of captured images according to age and gender. Parameters such as key size, input size, time taken, simulation, memory requirement, CPU usage. Major issues in multimodal biometrics such as matching algorithm, time delay, FAR are also discussed. Keywords Cryptographic algorithm · Advanced encryption standard · Biometric traits · FAR, FRR · Data encryption standard · Cipher text
S. S. More (B) MATS School of Information Technology, MATS University, Raipur, India e-mail: [email protected] B. Narain MATS University, Raipur, India e-mail: [email protected] B. T. Jadhav RIRD, Satara, Maharashtra, India e-mail: [email protected] © Springer Nature Singapore Pte Ltd. 2021 A. A. Rizvanov et al. (eds.), Advances in Biomedical Engineering and Technology, Lecture Notes in Bioengineering, https://doi.org/10.1007/978-981-15-6329-4_7
53
54
S. S. More et al.
1 Introduction Biometrics is used in computer science as a form of identification and access control. It is also used to identify individuals in groups (Shinde and Shinde 2014). The biometric process always goes into two modes i.e., Enrollment, Verification (Authentication)/Identification (Recognition) modes. The biometric implementation process consists of a sensor, feature extraction, matcher, and system database. Biometrics is becoming an important international standard as an authentication technology providing security controls in every field. Biometrics involves the use of physiological and behavioral characteristics to provide the identification of individuals as applied to physical and network security within a business. All the types of a biometric system are passes to these modes. If the first time an individual uses a biometric system then it is called the enrollment process and then biometric information is captured and stored. And Second time it is verified or recognized. In the verification or authentication mode, the system performs a one-to-one comparison of a captured biometric with a specific template stored in a biometric database to verify the individual is the person they claim to be. In recognition or identification mode the system performs a one-to-many comparison against a biometric database in an attempt to establish the identity of an unknown individual. There are lots of technical differences between every biometric type e.g., in the iris recognition system we analyze features of a colored ring of the eye and shape of the eyes, eyebrows, nose, lips, etc., measured in Facial recognition system, etc. (More et al. 2017). Instead of all the types of biometric systems, we captured images of fingerprint, face, and iris and palm print recognition. We apply a feature extraction algorithm on that images and applying the advanced cryptographic algorithm on that captured images by using MATLAB tool for analyzing the parameters such as key size, input size, time taken, simulation, memory requirement, CPU usage. (1) Fingerprint Recognition System: In the fingerprint recognition system we check the uniqueness of fingerprints is due to the series of ridges and furrows on the fingers. We also consider the arch, loop, and whorl of the captured fingerprint. For the fingerprint recognition system, we can use a previously defined feature extraction algorithm i.e., Canny Edge Detection/Gaussian Filter/Gaussian mixture/Gabor Filter Algorithm. (2) Face Recognition System: Facial recognition techniques measure facial characteristics. It measures the position, shape, and size of facial features such as eyebrows, eyes, nose, lips, and chin (More et al. 2017). In Face Recognition System we can use a previously defined feature extraction algorithm i.e., Gabor Filter Algorithm. (3) Iris Recognition System: The iris recognition technique analyzing feature of a colored ring of the eye (More and Jadahv 2016). Gabor filter algorithm is used for feature extraction of Iris recognition.
Advanced Encryption Standard Algorithm in Multimodal …
55
(4) Palmprint Recognition System: To determine the geometry features, principal line features, and wrinkle features we use processing techniques and applying these on the captured image (Sree Rama Murthy kora et al. 2012). For feature extraction of palmprint recognition, we can use Gabor Filter for texture analysis. Figure 1 shows sample images of different biometric traits.
Fig. 1 a Fingerprint, b face, c iris, d palmprint
Fig. 2 Definitions of palm print: principal lines (I-heart line, 2-head line, 3-life line, regions (Ifinger-root region, II-inside region, and III- outside region) and datum points. (a, b-endpoints, o-their midpoints)
56
S. S. More et al.
Fig. 3 Geometry features and delta points features of a palm print, where a–d is the perpendicular bisector of segment a–b, and points 1–5 are delta points
2 Multimodal Biometrics Multimodal biometrics is based on a combination of more than one type of biometric modalities or qualities. The advantages of a multimodal biometric system are higher accuracy, security, universality, and cost-effectiveness, etc. The goal of multimodal biometrics is used to reduce the biometric parametric errors i.e., False Accept Rate (FAR), False Reject Rate (FRR), and Failure to Enroll Rate (FTE). Multimodal Biometric System Fusion can be done by four levels and these are Sensor Level, Feature Level, Matching Score Level, and Decision Level. In sensor Level, biometric characteristics are coming from sensor level. In Feature Level fusion, the signal coming from different biometric channels is first proposed and feature vectors are extracted separately. In Matching Score Level combining the feature, we process them separately and the individual matching score is found. And lastly at decision Level, each modality is first pre-classified independently. The final classification is based on the fusion of the output of the different modalities. Multibiometric system may be Multi-algorithmic, Multi-instance, Multi-sensorial. We use this image by era (More and Jadhav 2017). Data encryption standard limitations are: 1. The key selected in the rounds is weak because during the splitting of keys to two half and swapping, it displays the same result if they have continuous 1s and 0s. 2. The S Box creates the same output from the different inputs on permutations. We can call these as Semi Weak keys. 3. If the message is encrypted with a particular key and it takes 1s complement of that encryption will be the same as that of the encryption of the compliment message and compliment key.
2.1 Role of Multimodal Biometrics in AES Cryptography is becoming an increasingly important feature of computer security (More et al. 2017). The proposed method gives the security to the whole system by
Advanced Encryption Standard Algorithm in Multimodal …
57
Fig. 4 Classification of a multimodal biometric system
using fingerprint, face, Palm, and facial features as a key in a cryptosystem. In the proposed model we use multimodal biometric features. Biometric template protection is one of the important issues in deploying a practical biometric system. To tackle this problem, many algorithms have been reported in recent years, most of them applying to use fingerprint, face, Palm, and face biometric. Since the contents and representation of every template are different than others. The template protection algorithm of one biometric trait cannot be directly applied to others. Moreover, we believe that no single template protection method can satisfy the diversity, revocability, security, and performance requirements (Bala and Joanna 2014). Data fusion is the process of integrating multiple data sources to produce more consistent, accurate, and useful information than that provided by any individual data source (Fig. 5). (a) AES Algorithm for Image Processing. The algorithm consists of combinations, permutations, and substitution between the images to be encrypted and the key is applied on both the encryption & decryption process. AES Algorithm is already developed we apply this on multimodal biometric images. (b) Tools Used For Digital Image Processing. MATLAB, Scilab, and Octave are widely used by engineers and scientists in both industry and academia for performing numerical computations and for developing and testing mathematical algorithms and image processing with related applications (Narain et al. 2013). In our work, we use the MATLAB tool for processing images, which is a high-performance language for technical computing. It is used in computation, visualization, and programming environment. (c) Implementations of AES Using Multimodal Biometric: Encrypted data set of sample images of fingerprint, face, Iris, and Palmprint are shown in Fig. 6.
58
S. S. More et al.
Fig. 5 Working on the AES algorithm
3 Methodology 3.1 Proposed Architecture The multimodal biometric system designed consists of six modules as in Fig. 1. • Fingerprint analysis module. • Iris analysis module.
Advanced Encryption Standard Algorithm in Multimodal …
a) Fingerprint-I1
b) Face-I2 Fig. 6 Encrypted data set of fingerprint, face, iris, palm print sample images
59
60
S. S. More et al.
c) Iris-I3
d) Palmprint-I4 Fig. 6 (continued)
Advanced Encryption Standard Algorithm in Multimodal …
61
Fig. 7 Decrypted dataset
• • • •
Palm print analysis module. Face analysis module Conversion and Fusion. AES Encryption/Decryption module.
The generation of secure biometric keys with the help of multimodal biometrics such as iris, fingerprint, face, and palm print is done as shown in Fig. 8.
3.2 Module Implementation The minutiae points are extracted from the fingerprint image, texture features from iris image, ROI score from palm print and shape of the eyes, eyebrows, nose, lips, etc.,
62
S. S. More et al.
Fig. 8 Generation of Secure biometric keys
measured in facial recognition system which is generated by the biometric system. The features are now converted into respective decimals. The decimals are converted into binaries and all the four binary follows XOR operation to generate the combined cryptographic key. The key is later compressed to Hexadecimal value which can act as the encryption key. The encryption key is now used for the DES encryption and decryption process. Again encryption and decryption is followed based on the two ciphers generated. Module 1: Biometrics Sensing. This module helps to recognize the biometric information of the users via sensors, cameras. • Images are generated which are further passed to module two for evaluation. • The Module helps to collect the information of human biometrics • The information to be collected are Fingerprint, Palm print, Iris, and Face. Module 2: Preprocessing Feature Extraction (Joseph and Parthiban 2016). • It helps to extract the features from human biometrics to generate biometric key • The features are extracted in the form of decimals which are then used to convert the binary value • Different techniques are followed for each biometric • At first, the image is enhanced, followed by thinning, segmentation.
Advanced Encryption Standard Algorithm in Multimodal …
63
Module 3: Normalization and Fusion Description (Joseph and Parthiban 2016) • This module helps to normalize the data or information gathered in the form of features to the type which can be used to create or generate the key. • The features are then fused by following the XOR operation of the biometric values obtained. Module 4: Generation of Keys Description • The above module and the current one are integrated to generate binary biometric ciphers key • The two binary keys generated are then considered as input to the next module of encryption. Module 5: AES Encryption and Decryption Description • The module follows the process of encryption and decryption by using the AES encryption process. • The binary keys generated from the above module are passed as inputs to generate encrypted ciphers.
4 Results and Discussion Analysis of about 5 different samples is followed and detailed evaluation can be seen from the table All the features are extracted and normalized in the form of binary values which later follows the proposed algorithm. All the values are passing as input to the proposed technique which provides better security due to 4 levels of multimodal biometrics (Isaac et al. 2018). This paper combines the scores based on the fusion of Iris, Fingerprint, face, and Palmprint data to generate biometric cryptographic keys. The analysis is done and it provides information about the performance and calculates approximately measures of the combined biometric techniques. The Iris, Fingerprint, face, and Palmprint data are collected from about 5 individuals and used for evaluation. Scores for each biometric trait are generated respectively. The calculation of analysis parameters such as key size, input size, time taken, simulation, memory requirement, CPU usage are estimated, and also FAR, FRR. The following figure shows the expected outcomes. In Fig. 9 we enter the ID of any person and then the preprocessing and enhancement process is done in Figs. 10 and 11. After that in Fig. 11 we apply Gabor and PCA filter for feature extraction to
Fig. 9 First step of entering input
64
S. S. More et al.
Fig. 10 Output of the program—preprocessing on the multimodal images
Fig. 11 Output of the program—enhancement process of fingerprint, face, iris, and palmprint
display a graphical representation of it in Fig. 12 and identifying or authenticate a person. The biometrics features for Iris, Fingerprint, face, and Palmprint are collected separately according to age and gender in the future. Then we apply the AES algorithm on these multimodal images.
Advanced Encryption Standard Algorithm in Multimodal …
65
Fig. 12 The output of the program—graphical representation of multimodal images
5 Conclusion In this paper, we have discussed the security issues of multimodal biometrics of the existing system which used the DES algorithm, but it has some limitations and this is overcome in the proposed system and gives more security in biometrics. The AES algorithm is already developed algorithm, here we apply it on the multimodal biometrics and checking the medium performance and then compare to the proposed system methodology as well as the algorithm by the experimental results and tabulations to justify the results. In the future, we will be trying to implement on a large dataset and for different cryptographic algorithms RSA, Blowfish, M-RSA, and proposed algorithm. This algorithm implementation time and matching algorithm’s time taken for completion should be minimized and try to avoid the false acceptance rate and false rejection rate in the system. And we will also try to compare the analysis of parameters of these algorithms with the proposed algorithm in the future. Ethical Clearance We are putting sample images since all the images shown in the study are images of one of the authors and no other participants were involved in the study. Hence, ethical approval was not required. Results shown are obtained using five samples (different orientations) of one of the authors of this paper. In the future, the study will be extended on a new dataset after obtaining approval of the institutional ethical committee with informed consent.
References Bala BK, Joanna JL (2014) Multi modal biometrics using cryptographic algorithm. Eur J Acad Essays 1(1):6–10 Isaac RA, Kathera A, Venkatachalam KH, Raj MT, Gokulnath G (2018) Spoofing Detection for fingerprint, palm-vein and facial recognition using deep representation 6(4):169–174
66
S. S. More et al.
Joseph T, Parthiban L (2016) Multimodal biometric based authentication for ensuring data security in Cloud Computing. J Chem Pharm Sci 9(4). ISSN: 0974-2115 More SS, Jadahv BT (2016) An overview on technologies used in biometric system. Int J Innov Res Comput Commun Eng 4(2):365–373 More SS, Jadhav BT (2017) Comparative study of biometric devices. Int J Innov Res Comput Commun Eng 5(2):1302–1309 More SS, Narain B, Jadahv BT (2017) A comparative analysis of unimodal and multimodal biometric systems. In: International conference on innovative trends in engineering science and management (ITESM-2017) Narain B, Zadgaonkar AS, Kumar S (2013) Impact of digital image processing on research and education. Natl Semin Work Shinde SJ, Shinde J (2014) Biometrics: overview and potential use for E-governance services. Int J Adv Res Comput Sci Softw Eng 4(6):1145–1151 Sree Rama Murthy kora, Verma P, Kashyap Y (2012) Palmprint recognition: palm print
In-Silico Construction of Hybrid ORF Protein to Enhance Algal Oil Content for Biofuel Mohit Nigam, Ruchi Yadav, and Garima Awasthi
Abstract As a renewable resource of biodiesel, algae has received global attention and hence our primary concern is to explore the new potential technologies for increasing yield of algal biofuel. The aim of our study was to develop a hybrid ORF, which will increase the oil producing capacity of algae on its expression as a functional protein. Utilizing BLAST and intense literature survey, 6 oil producing algal genes were selected which belonged to different superfamilies (NADB_Rossmann superfamily, RfaB superfamily, Aldo_ket_red superfamily, PP-binding Superfamily, LPLAT superfamily, and Acyl-ACP TE superfamily). Further, using the conserved regions of these superfamilies, the hybrid ORF was constructed by vector NTI tool. Since protein structure is an important aspect to verify protein sequence for its stability and existence in nature, multi-template homology modeling was done using Schrödinger software suite version 10.4.018. The five templates (1i24_A, 1pz1_A, 2jjm_A, 2m5r_A, 5 × 04_A) showing highest similarity with the hybrid ORF were selected for multi-template homology modeling. The comparative threedimensional structure prediction of the constructed hybrid ORF was done using Phyre2 server which is based on the principles of ab initio method. Structure verification of both the resulting models i.e., model generated by Schrödinger software and model predicted by Phyre2 server is done using Ramachandran plot and identified that functional region has been modeled accurately having 87.3% and 91% residues in allowed region, respectively. The metabolic pathway analysis of all the selected genes also verified, that they all are involved in lipid biosynthesis. This shows that the hybrid ORF designed can be potential tool to increase oil content in
M. Nigam · R. Yadav · G. Awasthi (B) Amity Institute of Biotechnology, Amity University Uttar Pradesh, Lucknow, UP, India e-mail: [email protected] M. Nigam e-mail: [email protected] R. Yadav e-mail: [email protected] © Springer Nature Singapore Pte Ltd. 2021 A. A. Rizvanov et al. (eds.), Advances in Biomedical Engineering and Technology, Lecture Notes in Bioengineering, https://doi.org/10.1007/978-981-15-6329-4_8
67
68
M. Nigam et al.
algae for biofuel production. Further in vitro analyses are required to study physiochemical properties of the protein and its effectiveness after constructing genetically engineered microorganism. Keywords Homology modeling · Schrödinger software · Phyre2 server · Microalgae · Biofuel
1 Introduction Biofuel research is currently an area of immense interest due to the increase in global energy demand. Biofuel is center of attention due to the emerging economies, increase in global oil prices, reduces environmental impact, foreign exchanges, and benefits to rural sector. Research on multiple approaches has been carried out currently for the use of microorganisms in the production of various types of feedstock for the biofuel production (Elshahed 2010; Demirbas 2009). Microalgae represents an exceptionally diverse but highly specialized group of microorganisms adapted to various ecological habitats. Algae is a good source for many products like carbohydrates, essential fatty acids, pigments, food supplements, fertilizer, pharmaceutical, and biofuel (Hemaiswarya et al. 2011). Many microalgae have the ability to produce high amounts of triacylglycerols (TAG) under photo-oxidative stress or different adverse environmental conditions that can be utilized as feedstock for producing biodiesel (Hu et al. 2008; Chisti 2007, 2008). Algae production is facing several challenges like water, nutrient, and contamination, which can overcome by using heterotropic and mixotropic algae. These algae are well known to grow faster and produce high oil content as feedstock for production biofuel compared to photoautotrophic cells (Miao and Wu 2004; Cerón García et al. 2000). The concept of strain improvement, leads to explore new ways based on genomics and transcriptomic information so as to improve the growth conditions of algae and increase lipid content that can help in reducing the cost of biofuel production. Many approaches are taken to enhance triacylglycerol (TAG) content in microalgae include over-expression of genes for biosynthesis of TAG, lipid catabolism inhibition, and interference with many other pathways (Radakovits et al. 2010). Genome engineering is gradually increasing its steps in algae research (Banerjee et al. 2018). In the present study in silico algal genes consortium is developed for expression of hybrid ORF protein to enhance the oil content in algae and will be fuel precursor molecules in TAG biosynthesis pathway.
In-Silico Construction of Hybrid ORF Protein …
69
2 Material and Methods Literature survey was done by journals and finally by NCBI genomic and proteomic databases for the identification of oil producing algal genes. 6 genes from different strains of algae were selected and its functional protein information was collected. Superfamily (Conserved regions) of these proteins was identified by BLASTp and then using these conserved regions of functional protein sequences, Hybrid ORF was constructed. Hybrid sequence verification and clone designing for the hybrid ORF was done using vector NTI tool (Guoqing and Moriyama 2004). Schrödinger software suite version 10.4.018 was used for multi-template homology modeling of the hybrid sequence. The 3D structure of hybrid protein was modeled and structure verification was done using Ramachandran plot (Schrödinger 2011, 2014; Ivanov et al. 2009; Ferrara and Edgar 2007). Ab initio method for protein structure prediction was also done by using Phyre2 server. The modeled structure was compared with the three-dimensional structure predicted using homology modeling method (Kelley et al. 2015). The best structure was identified for further analysis. The metabolic pathways of the selected algal genes were studied using KEGG database. Pathway study was done to identify the function and molecular mechanism of selected genes.
3 Result and Discussion 3.1 Identification of Algal Gene and Its Functional Protein Oil producing algal gene was searched through intensive literature survey and NCBI genomic and proteomic databases. Total of 6 genes and their functional proteins were identified and screened by literature survey as mentioned in Table 1. The 5 gene SQD1, SQD2, ACP1, CGLD24, and FAT1 were reported in Chlamydomonas reinhardtii only F751_2396 was identified in Auxenochlorella protothecoides. The protein sequences were saved in FASTA format and functional protein information was retrieved.
3.2 Hybrid ORF Construction Using Selected Conserved Regions of Superfamilies Protein screening of selected gene was done on the basis of different superfamilies, using BLASTp tool. The 6 target genes were further classified according to their superfamilies to which they belong, that were NADB_Rossmann superfamily, RfaB
70
M. Nigam et al.
Table 1 Algal oil producing genes and its functional protein information S. no.
Gene symbol
Protein
Organisms
Accession no.
1.
SQD1
UDP-sulfoquinovose synthase
Chlamydomonas reinhardtii
XP_001697898.1
2.
SQD2
Sulfolipid synthase
Chlamydomonas reinhardtii
XP_001689662.1
3.
F751_2396
D-arabinose 1-dehydrogenase
Auxenochlorella protothecoides
XP_011397567.1
4.
ACP1
Acyl-carrier protein
Chlamydomonas reinhardtii
XP_001699275.1
5.
CGLD24
Diacylglycerol acyl transferase
Chlamydomonas reinhardtii
XP_0016931C9.1
6.
FAT1
Acyl carrier protein thioesterase
Chlamydomonas reinhardtii
XP_001696619.1
superfamily, Aldo_ket_red superfamily, PP-binding Superfamily, LPLAT superfamily, and Acyl-ACP TE superfamily. Conserved sequences of these superfamilies were used for the designing of hybrid ORF using vector NTI tool. The selected superfamilies were studied to identify ORF start-stop positions and length of ORF, which was used for designing hybrid ORF as mentioned in Table 2. In silico techniques were used to study properties of hybrid sequence and its stability for cloning and expression in vector.
3.3 Hybrid ORF Clone Designed Using Vector NTI Tool Kit Vector NTI tool was used for the clone of designed hybrid ORF, also analysis and verification of cloned sequence was done. The length of hybrid ORF was 5175 bp and the gene sequence of hybrid ORF is mentioned below. Complete Sequence of Constructed Hybrid ORF Designed for the Biosynthesis of Lipid in Algal Strain. >hy_ORF _ length_5175 bp
Organisms
Chlamydomonas reinhardtii
S. no.
1.
NADB_Rossmann superfamily
Super family
UDP-sulfoquinovose synthase
Protein
STVRQATSSVRAASR ATSVKVQATPATVE KATAPAGSLSSNGA GTRVMIIGGDGYCG WATALHLSARGYEV CIVDNLCRRQFDLQL GLDTLTPIATIHDRV RRWGEVSGKHISLQI GDICDWEFLSQAFTS FKPNHVVHFGEQRS APYSMIDRQKAVFT QHNNVIGTINVLFAI KELQPDCHMVKLGT MGEYGTPNIDIEEGY ITINHNGRTDTLPYP KQGNSFYHLSKIHDS TNMLFTCKAWKIAA TDLNQGVVYGVRTD ETMADPLLLNRYDY DGIFGTALNRFVVQ AAVGHPLTVYGKGG QTRGFLDIRDTVRCI QLAIDNPAPKGEMR VYNQFTEQFSVNQL AEIVEREGKKLGLNV EVTKVPNPRVELEEH YYNAKCTKLRDLGL QPHLLADSMIDSLLE FAVTYKDRVRHELIK PAVDWRKTGVKVN TMGAAV
Conserved region sequence
Table 2 Selected conserved regions of superfamilies for designed hybrid ORF 40
ORF start position 477
ORF stop position 439
Length of ORF
(continued)
XP_001697898.1
Accession no.
In-Silico Construction of Hybrid ORF Protein … 71
Organisms
Aureococcus anophagefferens
S. no.
2.
Table 2 (continued)
RfaB superfamily
Super family
UDP-sulfoquinovose:DAG sulfoquinovosyltransferase
Protein
RPSKLPTTACRAAAT SDTDTPQRKVALLV EPTPFTHVSGYANRF QEMLKHLERRGDVV AVATPDDVPEAPASF GKFAVTTLGGFRFRP WYPEICLSLDLDGAA LQMIRDLDPDVVHA SSPGFLAVAALRRA GQGAERKPLLLSYH THIPVYVRKYASWV PFIEKTTWALLRAVH NRADLTIATSPQIRD ELLANGVTAIERVGV WNKGIDTDRFHPKF RSDAARARMTSGHP GDKLAVYVGRLGVE KRIDELRGVLEAIPEL RLALVGAGPAEPGL RETFADVADRVVFT GLLRGDELSAAFAS ADVFLMPSDSETLGF VVLESMASGVPVVG CRAGGIPNLIDDDQE GATGRLHAVGDVAE IAELTRGLLDDAPKR DAMGAAARAEAER WDWASSGETLRADS YGAAIRNFAAR
Conserved region sequence 22
ORF start position 430
ORF stop position 415
Length of ORF
(continued)
XP_009034698.1
Accession no.
72 M. Nigam et al.
Organisms
Auxenochlorella protothecoides
S. no.
3.
Table 2 (continued)
Aldo_ket_red superfamily
Super family
L-galactose dehydrogenase
Protein
RPLGSTGLEISIIGFG ASPLGNVFGDVHQD TATEAVRTAFDLGIT LFDTSPFYGLTKSED VLGQALRDAGLPRD QFVLATKVGRYGQD TFDFSGPRVTRSVEE SLERLHTSYIDLIQVH DMEFGSLDQIIAETL PALQRLKEKGLVRHI GITGLPLACFQYVLD RVPCGTVDVVLSYC HYTLCDQSLGRILPY LESKAVGVINASVLG MGLLTPHGPPAWHP APAELQAAARAAAR AADAHAVDLPKLAT MFSVANPGIATHLIG FSTPDQVRNAVHAV LQAQGLEENAQAEQ EERAMVEIREILAGT AQVTWPSGLPEN
Conserved region sequence 8
ORF start position 327
ORF stop position 320
Length of ORF
(continued)
XP_011397567.1
Accession no.
In-Silico Construction of Hybrid ORF Protein … 73
Organisms
Chlamydomonas reinhardtii
Chlamydomonas reinhardtii
S. no.
4.
5.
Table 2 (continued)
LPLAT superfamily
PP-binding superfamily
Super family
diacylglycerol acyl transferase
acyl-carrier protein
Protein
AAAYFPTRVVVTDP EAFRTDRGYLFGFCP HSALPIALPIAFATTA SPLLPKELRGRTHGL ASSVCFSAPIVRQLY WWLGVRPATRQSIS GLLDKCRARKVAVL VPGGVQEVLNMEHG KEVAYLSSRTGFVRL AVQHGAPLVPVWAF GQTRAYSWFRPGPP LVPTWLVERISRAAG AVPIIFHGRGMFGQY GTPLMPHREPLTIVV GRPIPVPELAPGQLE PEPEVLAALLKRFTD DLQALYDKHKAQFG VPKGEELVI
KDSVTERVLHVTKH FEKIDASKVSPAASF EKDLGLDSLDVVEL VMALEEEFGLEIPDA EADKIASVGDAINYI CSNP
Conserved region sequence
81
49
ORF start position
326
125
ORF stop position
258
77
Length of ORF
(continued)
XP_001693189.1
XP_001699275.1
Accession no.
74 M. Nigam et al.
Organisms
Chlamydomonas reinhardtii
S. no.
6.
Table 2 (continued)
Acyl-ACP TE superfamily
Super family
acyl carrier protein thioesterase
Protein
SFREEHRIRGYEVSP DQRATIVTVANLLQE VAGNHAVGMWGRT DEGFASLPSMKDYN LLFVMTRLQVRMYE YPKWGDVVAVETYF TEEGRLAFRREWKL MDVATGKLLGAGTS TWVTINTATRRLSKL PEDVRKRFLRFAPPS SVHILPPEETKKKLQ DMPKYELPGQVQSA QQVARRADMDMNG HINNVTYLAWTLESL PERVMSGGYKMQEI ELDFKAECTAGNAIE AHCNP
Conserved region sequence 90
ORF start position 318
ORF stop position 234
Length of ORF
XP_001696619.1
Accession no.
In-Silico Construction of Hybrid ORF Protein … 75
76
M. Nigam et al.
AGCACCGTGCGCCAGGCCACCAGCAGCGTGCGCGCCGCCAGCCGCGCCACCAGCGTGA AGGTGCAGGCCACCCCCGCCACCGTGGAGAAGGCCACCGCCCCCGCCGGCAGCCTGAG CAGCAACGGCGCCGGCACCCGCGTGATGATCATCGGCGGCGACGGCTACTGCGGCTGG GCCACCGCCCTGCACCTGAGCGCCCGCGGCTACGAGGTGTGCATCGTGGACAACCTGT GCCGCCGCCAGTTCGACCTGCAGCTGGGCCTGGACACCCTGACCCCCATCGCCACCATC CACGACCGCGTGCGCCGCTGGGGCGAGGTGAGCGGCAAGCACATCAGCCTGCAGATCG GCGACATCTGCGACTGGGAGTTCCTGAGCCAGGCCTTCACCAGCTTCAAGCCCAACCAC GTGGTGCACTTCGGCGAGCAGCGCAGCGCCCCCTACAGCATGATCGACCGCCAGAAGG CCGTGTTCACCCAGCACAACAACGTGATCGGCACCATCAACGTGCTGTTCGCCATCAAG GAGCTGCAGCCCGACTGCCACATGGTGAAGCTGGGCACCATGGGCGAGTACGGCACCC CCAACATCGACATCGAGGAGGGCTACATCACCATCAACCACAACGGCCGCACCGACAC CCTGCCCTACCCCAAGCAGGGCAACAGCTTCTACCACCTGAGCAAGATCCACGACAGC ACCAACATGCTGTTCACCTGCAAGGCCTGGAAGATCGCCGCCACCGACCTGAACCAGG GCGTGGTGTACGGCGTGCGCACGACGAGACCATGGCCGACCCCCTGCTGCTGAACCGC TACGACTACGACGGCATCTTCGGCACCGCCCTGAACCGCTTCGTGGTGCAGGCCGCCGT GGGCCACCCCCTGACCGTGTACGGCAAGGGCGGCCAGACCCGCGGCTTCCTGGACATC CGCGACACCGTGCGCTGCATCCAGCTGGCCATCGACAACCCCGCCCCCAAGGGCGAGA TGCGCGTGTACAACCAGTTCACCGAGCAGTTCAGCGTGAACCAGCTGGCCGAGATCGT GGAGCGCGAGGGCAAGAAGCTGGGCCTGAACGTGGAGGTGACCAAGGTGCCCAACCC CCGCGTGGAGCTGGAGGAGCACTACTACAACGCCAAGTGCACCAAGCTGCGCGACCTG GGCCTGCAGCCCCACCTGCTGGCCGACAGCATGATCGACAGCCTGCTGGAGTTCGCCGT GACCTACAAGGACCGCGTGCGCCACGAGCTGATCAAGCCCGCCGTGGACTGGCGCAAG ACCGGCGTGAAGGTGAACACCATGGGCGCCGCCGTGATGATCCGCTACCTGGTGGAGG CCGGCTGCCAGGTGCTGGTGGTGACCACCGGCGCCGGCTACACCCTGCCCGGCGTGGA CGCCAGCAGCTTCCGCGAGCAGCCCGAGACCTTCGCCGGCGCCCGCGTGGTGAGCGCC CTGAGCTTCGGCTGCCCCTGGTACCTGCAGGTGCCCCTGACCTTCGCCCTGAGCCCCCG CATCTGGCGCGAGGTGCGCGACTTCCAGCCCGACCTGATCCACTGCAGCAGCCCCGGC GTGATGGTGTTCGCCGCCAAGCTGTACGCCTGGCTGCTGAAGAAGCCCATCGTGCTGAG
In-Silico Construction of Hybrid ORF Protein …
77
CTACCACACCCACGTGCCCAGCTACCTGCCCAAGTACGGCATCAGCTACCTGGTGCCCG CCATGTGGGGCTTCCTGCGCATGCTGCACGCCACCGCCCACCTGACCCTGACCGTGAGC CCCGCCATGGTGGACGAGCTGGTGACCAACCGCGCCGTGAACAGCGCCAAGCAGGTGC AGGTGTGGAAGAAGGGCGTGGACAGCGAGGTGTTCCACCCCCGCTTCCGCAGCGCCGC CATGCGCGAGCGCCTGACCGGCGGCCAGCCCGACCGCCCCACCCTGGTGTACGTGGGC CGCCTGGGCTTCGAGAAGAACCTGTTCTTCCTGCGCGAGGTGCTGCAGCGCAACCCCGG CGTGGGCCTGGCCTTCGTGGGCGACGGCCCCGCCCGCCAGGAGCTGCAGGCCGCCTTC AAGGGCACCCCCACCCAGTTCCTGGGCATGCTGCACGGCGAGGACCTGAGCGCCGCCT ACGCCAGCAGCGACATCTTCGTGATGCCCAGCGAGAGCGAGACCCTGGGCTTCGTGGT GCTGGAGGCCATGGCCAGCGAGCTGCCCGTGGTGGCCGTGCGCGCCGGCGGCATCCCC GACATCATCACCCCCGGCGACAGCGGCGTGACCGGCTTCCTGTACGAGCCCGGCGACG TGGACAAGGCCGCCGAGCTGATCCAGCAGCTGGCCGCCGACGCCCAGCTGCGCAGCCG CGTGGGCATCCGCGCCCGCCAGGAGGTGGCCAAGTGGGACTGGCGCGCCGCCACCATG CACCTGCTGAACGTGCAGTACCCCATCGCCATGGCCGCCGCCGCCGCCCAGTACGGCG AGGCCCTGGGCCGCGTGCAGTGGCTGCCCGCCCAGGACGCCCTGGCCGCCCAGCCCCC CCAGCGCCCCCTGGGCAGCACCGGCCTGGAGATCAGCATCATCGGCTTCGGCGCCAGC CCCCTGGGCAACGTGTTCGGCGACGTGCACCAGGACACCGCCACCGAGGCCGTGCGCA CCGCCTTCGACCTGGGCATCACCCTGTTCGACACCAGCCCCTTCTACGGCCTGACCAAG AGCGAGGACGTGCTGGGCCAGGCCCTGCGCGACGCCGGCCTGCCCCGCGACCAGTTCG TGCTGGCCACCAAGGTGGGCCGCTACGGCCAGGACACCTTCGACTTCAGCGGCCCCCG CGTGACCCGCAGCGTGGAGGAGAGCCTGGAGCGCCTGCACACCAGCTACATCGACCTG ATCCAGGTGCACGACATGGAGTTCGGCAGCCTGGACCAGATCATCGCCGAGACCCTGC CCGCCCTGCAGCGCCTGAAGGAGAAGGGCCTGGTGCGCCACATCGGCATCACCGGCCT GCCCCTGGCCTGCTTCCAGTACGTGCTGGACCGCGTGCCCTGCGGCACCGTGGACGTGG TGCTGAGCTACTGCCACTACACCCTGTGCGACCAGAGCCTGGGCCGCATCCTGCCCTAC CTGGAGAGCAAGGCCGTGGGCGTGATCAACGCCAGCGTGCTGGGCATGGGCCTGCTGA CCCCCCACGGCCCCCCCGCCTGGCACCCCGCCCCCGCCGAGCTGCAGGCCGCCGCCCGC GCCGCCGCCCGCGCCGCCGACGCCCACGCCGTGGACCTGCCCAAGCTGGCCACCATGT TCAGCGTGGCCAACCCCGGCATCGCCACCCACCTGATCGGCTTCAGCACCCCCGACCAG GTGCGCAACGCCGTGCACGCCGTGCTGCAGGCCCAGGGCCTGGAGGAGAACGCCCAGG CCGAGCAGGAGGAGCGCGCCATGGTGGAGATCCGCGAGATCCTGGCCGGCACCGCCCA GGTGACCTGGCCCAGCGGCCTGCCCGAGAACAAGGACAGCGTGACCGAGCGCGTGCTG CACGTGACCAAGCACTTCGAGAAGATCGACGCCAGCAAGGTGAGCCCCGCCGCCAGCT TCGAGAAGGACCTGGGCCTGGACAGCCTGGACGTGGTGGAGCTGGTGATGGCCCTGGA GGAGGAGTTCGGCCTGGAGATCCCCGACGCCGAGGCCGACAAGATCGCCAGCGTGGGC GACGCCATCAACTACATCTGCAGCAACCCCGCCGCCGCCTACTTCCCCACCCGCGTGGT GGTGACCGACCCCGAGGCCTTCCGCACCGACCGCGGCTACCTGTTCGGCTTCTGCCCCC ACAGCGCCCTGCCCATCGCCCTGCCCATCGCCTTCGCCACCACCGCCAGCCCCCTGCTG CCCAAGGAGCTGCGCGGCCGCACCCACGGCCTGGCCAGCAGCGTGTGCTTCAGCGCCC CCATCGTGCGCCAGCTGTACTGGTGGCTGGGCGTGCGCCCCGCCACCCGCCAGAGCATC AGCGGCCTGCTGGACAAGTGCCGCGCCCGCAAGGTGGCCGTGCTGGTGCCCGGCGGCG
78
M. Nigam et al.
TGCAGGAGGTGCTGAACATGGAGCACGGCAAGGAGGTGGCCTACCTGAGCAGCCGCAC CGGCTTCGTGCGCCTGGCCGTGCAGCACGGCGCCCCCCTGGTGCCCGTGTGGGCCTTCG GCCAGACCCGCGCCTACAGCTGGTTCCGCCCCGGCCCCCCCCTGGTGCCCACCTGGCTG GTGGAGCGCATCAGCCGCGCCGCCGGCGCCGTGCCCATCATCTTCCACGGCCGCGGCA TGTTCGGCCAGTACGGCACCCCCCTGATGCCCCACCGCGAGCCCCTGACCATCGTGGTG GGCCGCCCCATCCCCGTGCCCGAGCTGGCCCCCGGCCAGCTGGAGCCCGAGCCCGAGG TGCTGGCCGCCCTGCTGAAGCGCTTCACCGACGACCTGCAGGCCCTGTACGACAAGCA CAAGGCCCAGTTCGGCGTGCCCAAGGGCGAGGAGCTGGTGATCAGCTTCCGCGAGGAG CACCGCATCCGCGGCTACGAGGTGAGCCCCGACCAGCGCGCCACCATCGTGACCGTGG CCAACCTGCTGCAGGAGGTGGCCGGCAACCACGCCGTGGGCATGTGGGGCCGCACCGA CGAGGGCTTCGCCAGCCTGCCCAGCATGAAGGACTACAACCTGCTGTTCGTGATGACCC GCCTGCAGGTGCGCATGTACGAGTACCCCAAGTGGGGCGACGTGGTGGCCGTGGAGAC CTACTTCACCGAGGAGGGCCGCCTGGCCTTCCGCCGCGAGTGGAAGCTGATGGACGTG GCCACCGGCAAGCTGCTGGGCGCCGGCACCAGCACCTGGGTGACCATCAACACCGCCA CCCGCCGCCTGAGCAAGCTGCCCGAGGACGTGCGCAAGCGCTTCCTGCGCTTCGCCCCC CCCAGCAGCGTGCACATCCTGCCCCCCGAGGAGACCAAGAAGAAGCTGCAGGACATGC CCAAGTACGAGCTGCCCGGCCAGGTGCAGAGCGCCCAGCAGGTGGCCCGCCGCGCCGA CATGGACATGAACGGCCACATCAACAACGTGACCTACCTGGCCTGGACCCTGGAGAGC CTGCCCGAGCGCGTGATGAGCGGCGGCTACAAGATGCAGGAGATCGAGCTGGACTTCA AGGCCGAGTGCACCGCCGGCAACGCCATCGAGGCCCACTGCAACCCC
Construction of Hybrid ORF and its Restriction Enzyme Analysis. Hybrid ORF was constructed showing 6 conserved regions of selected superfamilies. These were further analyzed for restriction enzyme analysis. The restriction sites were identified in hybrid DNA sequence for many restriction enzymes as shown in Fig. 1. These restriction enzymes and their sites help in construction of hybrid ORF for further in silico or in vitro cloning and analysis.
Fig. 1 Restriction enzyme mapping of constructed hybrid ORF
In-Silico Construction of Hybrid ORF Protein …
79
Identification and Analysis of Hybrid ORF using ORF Finder. The hybrid ORF identification was done using ORF finder. The hybrid ORFs are marked with directional arrow in the sequence panel. According to the observation the arrows were solid, which display complete ORFs. As the analysis was done for complete ORF, which was run on default parameters and the minimum size base pair was specified as 150. The start codons were ATG GTG and stop codons were TAA TGA TAG for the ORF finder analysis. The positions of ORFs were marked in the sequence, which is shown in Fig. 2. The ORF finder also translates ORF in amino acid sequence, which is visible in Fig. 3.
Fig. 2 Identification and analysis of ORF in hybrid sequence using ORF Finder
Fig. 3 Thermodynamics analysis of hybrid ORF as predicted from vector NTI tool
80 Table 3 Thermodynamics analysis for the Hybrid ORF on various parameters
M. Nigam et al. S. no.
Parameters of thermodynamics analysis
Results
1.
dG Temperature(C)
25.0
2.
Probe Conc. (pMol)
250.0
3.
Salt Conc. (mMol)
50.0
4.
% Formamide
0.0
5.
Therm. Tm.
100.0
6.
GC Content
67.6%
7.
%GC Tm
87.0
8.
Stem Length (bp)
3
9.
Palindromes (bp)
6
10.
3’ End dG
−18.2
11.
3’ End Length (bp)
7
12.
Nucl. Repeats (bp)
4
13.
Mol. Wt.
307492.1
14.
dH
−8640.2
15.
dG
−2322.1
16.
dS
−21184.8
Thermodynamics Analysis of Hybrid ORF. The thermodynamic analysis was done to verify the hybrid clone as well as to check the stability by studying important thermodynamics parameters as shown in Fig. 3 (Table 3). Vector NTI® Express designer calculates two different melting temperatures for DNA/RNA oligonucleotides, which is thermodynamic Tm (Therm. Tm) and %GC Tm. The analysis was carried out at default parameters like probe concentration, 250 pM and salt concentration, 50 mM for thermodynamic calculation. The GC content of designed hybrid ORF was 67.6% and % GC Tm was 87.0, which was higher. GC content higher than 60% is considered for the gene design, protein expression, and primer design for PCR experiments. The GC content is important because the GC pair is having three hydrogen bonds, whereas AT pairs are having only two hydrogen bonds, which affect the stability of DNA. The GC content also affect the secondary structure of mRNA and annealing temperature in PCR experiments for template DNA (BIC, Homepage, GC Calculator 2018). Cloned Hybrid ORF in Invitrogen Vector pENTR/D-TOPO of length 2580 base pair (circular form). The selected 6 genes were cloned as shown in Fig. 4, into invitrogen vector pENTR/D-TOPO of length 2580 bp as the small vector is preferred for the contraction of multigene sequence cloning and it also provides stability to the designed clone. Vector NTI tool constructed the desired clone for further verification.
In-Silico Construction of Hybrid ORF Protein …
81
Fig. 4 Cloned Hybrid ORF in vector pENTR/D-TOPO (circular form) using TOPO cloning method
3.4 Multi-template Homology Modeling of Hybrid ORF Protein Using Schrödinger Software Hybrid ORF construction was done and the hybrid protein structure was modeled using Schrödinger software based on multi-template homology modeling method. 1i24_A, 1pz1_A, 2jjm_A, 2m5r_A, 5 × 04_A, are the 5 selected templates which were showing similarity with the hybrid ORF sequence and were used to model the hybrid protein as shown in Fig. 5.
Fig. 5 Protein structure of hybrid ORF. Protein structure prediction was done using multi-template homology modeling method as included in Schrödinger software
82
M. Nigam et al.
Fig. 6 Ramachandran plot of hybrid protein structure using Schrödinger software showing 87.3% of residues in favored and allowed regions and 12.7% in disallowed region
Structure verification was done using Ramachandran plot (Ramachandran et al. 1963; Kleywegt and Jones 1996; Lovell et al. 2003; Ho and Brasseur 2005; Wang et al. 2009) as shown in Fig. 6, showing 87.3% of residues in favored and allowed regions and 12.7% in disallowed region. The protein structure obtained from multi-template homology modeling using Schrödinger software was not showing appropriate results, so the ab initio method was used for protein structure prediction using Phyre2 server.
3.5 Protein Structure Prediction of the Constructed Hybrid ORF by Phyre2 Server Phyre2 server was used for the comparative 3-D structure prediction of the constructed hybrid ORF sequence, which is based on ab initio method. A comparatively better model was predicted using Phyre2 server and protein structure was shown in Fig. 7. The three dimensional structures of proteins are predicted, on the basis of amino acid sequence by using ab initio method. Structure verification of predicted protein was done using Ramachandran plot as shown in Fig. 8. Ramachandran plot shows that 91% of residues lie in favored region (red area) and 9% in allowed region (yellow area). This Ramachandran plot shows secondary structure like α helix and β sheet in the fully allowed part (favored region) and outer limit (allowed region). The generously allowed regions are located as pale greenish yellowish area. The disallowed regions (white area) generally involve
In-Silico Construction of Hybrid ORF Protein …
83
Fig. 7 Protein structure of hybrid ORF build using ab initio method by Phyre2 server
Fig. 8 Ramachandran plot of predicted 3-D protein model using Phyre2 server showing 91% of residues in favored region and 9% in allowed regions
steric hindrance between the side chains of one amino acid with the backbone of the succeeding amino acid. The glycine is an exception since it lacks the side chain responsible for the clash and can adopt phi and psi angles in disallowed region of the Ramachandran plot, which is acceptable (Cragg and Newman 2013).
84
M. Nigam et al.
Metabolic Pathway Analysis of Algal Genes Selected for Designing of Hybrid ORF. These selected genes are involved in glycerolipid metabolism and metabolic pathways, which are supporting fatty acid biosynthesis. These genes are involved in lipid biosynthesis (SQD1 and SQD2), glucose catabolism (F751_2396), enhance the lipid content even in stressed condition (CGLD24), support growth of fatty acid chain (ACP1), and increases lipid synthesis through protein–protein interaction (FAT1). These selected genes favor lipid biosynthesis during its metabolic pathway and are not involved in any other biosynthesis pathway except ACP1 and CGLD24. Acylcarrier protein of ACP1 gene is also involved in polyketide secondary metabolite biosynthesis. Polyketides are medically important compounds, used as antibiotics, anticancer agents, immune suppressants, and cholesterol-lowering agents (SWISSMODEL | Course | Secondary Structure and Backbone Conformation 2018). DGAT protein of CGLD24 is involved in triacylglycerol metabolism in higher eukaryotes for physiologic processes like intestinal fat absorption, lipoprotein assembly, etc. (Wang et al. 2007). Almost the other function of ACP1 and CGLD24 gene deals with the fatty components. The details of all the selected gene regarding the metabolic pathway is mentioned in Table 4. Mechanism of Selected Gene during Metabolic Pathway. Lipid biosynthesis is having two steps; the first step reaction is catalyzed by SQD1 protein in which UDPsulfoquinovose is assembled from UDP-glucose and sulfite. The second step SQD2 protein catalysis the sulfoquinovose from UDP-sulfoquinovose to diacylglycerol (Sanda et al. 2001; Yu et al. 2002). D-arabinose 1-dehydrogenase catalyzes Darabinose into D-arabinono-1,4-lactone, in heterotrophic conditions this protein helps in synthesis of fatty acids (Gao et al. 2014). Acyl carrier protein (ACP) is a conserved carrier of acyl intermediates and provides substrates to enzymes, which are responsible for fatty acid biosynthesis (Byers and Gong 2007). Diacylglycerol acyltransferase (DGATs) catalyze the formation of diacylglycerol from 2monoacylglycerol and fatty acyl-CoA. DGATs also catalyze the terminal step in triacylglycerol synthesis by using diacylglycerol and fatty acyl-CoA as substrates. DGAT has two isoforms DGAT1 and DGAT2, which is reported in plant and animals. The deficiency of DGATs accumulates less triacylglycerol (Zou et al. 1999; Smith et al. 2000; Stone et al. 2004), whereas triacylglycerol increases, as DGAT enzyme over-expression increases in plants (Andrianov et al. 2010; Xu et al. 2008), animals (Kamisaka et al. 2010; Liu et al. 2009) and yeast (Kamisaka et al. 2007). It has been found that most algae have multiple copies of putative DGAT2s genes, whereas single genes are reported in other eukaryotes (Chen and Smith 2012). It has been reported earlier that microalgae are capable of producing triacylglycerols (TAG) in stress condition. Phospholipid: Diacylglycerol acyltransferase (PDAT) catalyzes TAG synthesis in green microalga Chlamydomonas reinhardtii through two pathways: transacylation of diacylglycerol (DAG) with acyl groups from phospholipids and galactolipids and DAG:DAG transacylation. PDAT also possesses acyl hydrolase activities using TAG, phospholipids, galactolipids, and cholesteryl esters as substrates (Kamisaka et al. 2007). The fatty acid acyl carrier protein (ACP) along with thioesterase (TE) undergoes protein–protein interactions and results in fatty acid
F751_2396 Auxenochlorella protothecoides
3.
D-arabinose 1-dehydrogenase
Catalyzes the oxidation of D-arabinose, L-xylose, L-fucose and L-galactose in the presence of NADP+
–
Catalyzes the – transfer of the sulfoquinovose moiety from UDP-sulfoquinovose to diacylglycerol during sulfolipid biosynthesis
SQD2
2.
Chlamydomonas Sulfolipid synthase reinhardtii
Chlamydomonas UDP-sulfoquinovose Converts – reinhardtii synthase UDP-glucose and sulfite to the sulfolipid head group precursor UDP-sulfoquinovose (Sulfolipid biosynthesis)
Apro 00053, apro 01100, apro 01110
Cre 00561, cre 01100
Cre 00561, cre 00520
References
Ascorbate and aldarate metabolism, Metabolic pathways, Biosynthesis of secondary metabolites
Glycerolipid metabolism, Metabolic pathways
(continued)
Gao et al. 2014)
Yu et al. 2002)
Glycerolipid Sanda et al. 2001) metabolism, Amino sugar and nucleotide sugar metabolism
Protein KEGG Name of function in any pathway pathway other id biosynthesis
SQD1
Protein function in lipid biosynthesis
1.
Protein
Micro organism
S. no. Gene symbol
Table 4 Metabolic pathways of genes responsible for algal oil production
In-Silico Construction of Hybrid ORF Protein … 85
CGLD24
FAT1
5.
6.
Chlamydomonas Acyl carrier protein reinhardtii thioesterase
Chlamydomonas Diacylglycerol acyl reinhardtii transferase
Chlamydomonas Acyl-carrier protein reinhardtii
ACP1
4.
Protein
Micro organism
S. no. Gene symbol
Table 4 (continued)
Plays an essential role in chain termination during de novo fatty acid synthesis
Transferring acyl groups other than amino-acyl groups (transferase activity)
Carrier of the growing fatty acid chain in fatty acid biosynthesis
Protein function in lipid biosynthesis
-
DGAT is important in higher eukaryotes for physiologic processes involving triacylglycerol metabolism such as intestinal fat absorption, lipoprotein assembly, adipose tissue formation, and lactation
Polyketide secondary metabolite biosynthesis
Cre 00061
Cre 00561, cre 01100
Cre 00190, cre 01100
References
Fatty acid biosynthesis
Glycerolipid metabolism, Metabolic pathways
Blatti et al. 2012)
Patent and https:, , patecm, patent, US9771605, last accessed 2018, 12, 14, 2018; Cases et al. 1998)
Oxidative Byers and Gong 2007) phosphorylation, Metabolic pathways
Protein KEGG Name of function in any pathway pathway other id biosynthesis
86 M. Nigam et al.
In-Silico Construction of Hybrid ORF Protein …
87
hydrolysis within the algal chloroplast (Blatti et al. 2012). The metabolic pathway analysis verifies that selected genes sequences for the construction of hybrid ORF protein are involved in lipid biosynthesis.
4 Conclusion The in silico construction of genetic engineered microorganism was done by constructing hybrid ORF which was capable of expressing functional protein to enhance oil content in algae. The hybrid ORF protein was modeled by Schrödinger software and Phyre2 server. Structural verification was done by Ramachandran plot, which shows that both the constructed hybrid protein models are stable, with 87.3% residues in favored region by Schrödinger software and 91% residues in favored region by Phyre2 server. These comparative results suggest that Phyre2 web-based homology modeling method is preferred over multi-template homology modeling done by Schrödinger software. The stability of constructed hybrid ORF protein was also verified by thermodynamic analysis, through which GC content estimated was 67.6%. The estimated GC content favors the stability of hybrid ORF protein. The metabolic pathway analysis also verified that all the selected genes used for designing hybrid ORF are majorly involved in lipid synthesis pathways. These results justify the construction of hybrid ORF protein, which will have the potential to serve as an effective tool for designing of genetically engineered microorganisms to enhance high cellular oil content in algae. This complete in silico analysis will be a great support for construction of genetically engineered microorganisms during in vitro studies. The wet laboratory studies will be conducted in near future for the in vitro verification of hybrid ORF, constructed to enhance oil content in algal strains. This study will provide an immense support in the area of biofuel research. Acknowledgements We like to acknowledge tools and databases for bioinformatics analysis present at Amity Institute of Biotechnology, Amity University Uttar Pradesh, Lucknow campus for conducting this study and also our immense gratitude and deep regard to all those who directly or indirectly helped us to successfully complete this work. This research project is not funded by any specific grant from funding agencies in the public, commercial, or nonprofit sectors.
References Andrianov V, Borisjuk N, Pogrebnyak N, Brinker A, Dixon J, Spitsin S, Flynn J, Matyszczuk P, Andryszak K, Laurelli M, Golovkin M, Koprowski H (2010) Tobacco as a production platform for biofuel: overexpression of Arabidopsis DGAT and LEC2 genes increases accumulation and shifts the composition of lipids in green biomass. Plant Biotechnol J 8:277–287 Banerjee A, Banerjee C, Negi S, Chang JS, Shukla P (2018) Improvements in algal lipid production: a systems biology and gene editing approach. Crit Rev Biotechnol 38(3):369–385
88
M. Nigam et al.
BIC, Homepage, GC Calculator (2018) https://www.biologicscorp.com/tools/GCContent/#.XBO U4NszbIU. Accessed 14 Dec 2018 Blatti JL, Beld J, Behnke CA, Mendez M, Mayfield SP, Burkart MD (2012) Manipulating fatty acid biosynthesis in microalgae for biofuel through protein-protein interactions. PLoS ONE 7(9):e42949. https://doi.org/10.1371/journal.pone.0042949 Byers DM, Gong H (2007) Acyl carrier protein: structure-function relationships in a conserved multifunctional protein family. Biochem Cell Biol. 85(6):62–649 Cases S, Smith SJ, Zheng YW, Myers HM, Lear SR, Sande E, Novak S, Collins C, Welch CB, Lusis AJ, Erickson SK, Farese RV Jr (1998) Identification of a gene encoding an acyl CoA:diacylglycerol acyltransferase, a key enzyme in triacylglycerol synthesis. Proc Natl Acad Sci 95(22):13018–13023 Cerón García MC, Fernández Sevilla JM, Acién Fernández FG, Grima EM, Camacho FG (2000) Mixotrophic growth of Phaeodactylum tricornutum on glycerol: growth rate and fatty acid profile. J Appl Phycol 12:239–248 Chen J, Smith AG (2012) A look at diacylglycerol acyltransferases (DGATs) in algae. J Biotechnol 162(1):28–39 Chisti Y (2007) Biodiesel from microalgae. Biotechnol Adv 25(3):294–306 Chisti Y (2008) Biodiesel from microalgae beats bioethanol. Trends Biotechnol 26(3):126–131 Cragg GM, Newman DJ (2013) Natural products: a continuing source of novel drug leads. Biochim Biophys Acta 1830(6):3670–3695 Demirbas A (2009) Political, economic and environmental impacts of biofuels: a review. Appl Energy 86:108–117 Elshahed MS (2010) Microbiological aspects of biofuel production: current status and future directions. J Adv Res 1(2):103–111 Ferrara P, Edgar J (2007) Evaluation of the utility of homology models in high throughput docking. J Mol Model 13:897–905 Gao C, Wang Y, Shen Y, Yan D, He X, Dai J, Wu Q (2014) Oil accumulation mechanisms of the oleaginous microalga Chlorella protothecoides revealed through its genome, transcriptomes, and proteomes. BMC Genomics 10:15–582 Guoqing L, Moriyama EN (2004) Vector NTI: a balanced all-in-one sequence analysis suite. Brief Bioinform 5:378–388 Hemaiswarya S, Raja R, Ravi Kumar R, Ganesan V, Anbazhagan C (2011) Microalgae: a sustainable feed source for aquaculture. World J Microbiol Biotechnol 27:1737–1746 Ho BK, Brasseur R (2005) The Ramachandran plots of glycine and pre-proline. BMC Struct Biol 5(14). doi.org/10.1186/1472–6807–5–14 Hu Q, Sommerfeld M, Jarvis E, Ghirardi M, Posewitz M, Seibert M, Darzins A (2008) Microalgal triacylglycerols as feedstocks for biofuel production: perspectives and advances. Plant J 54:621– 639 Ivanov AA, Barak D, Jacobson AK (2009) Evaluation of homology modeling of G proteincoupled receptors in light of the A2A adenosine receptor crystallographic structure. J Med Chem 52(10):3284–3292 Kamisaka Y, Tomita N, Kimura K, Kainou K, Uemura H (2007) DGA1 (diacylglycerol acyltransferase gene) overexpression and leucine biosynthesis significantly increase lipid accumulation in the Deltasnf2 disruptant of Saccharomyces cerevisiae. Biochem J 408:61–68 Kamisaka Y, Kimura K, Uemura H, Shibakami M (2010) Activation of diacylglycerol acyltransferase expressed in Saccharomyces cerevisiae: overexpression of Dga1p lacking the N-terminal region in the Deltasnf2 disruptant produces a significant increase in its enzyme activity. Appl Microbiol Biotechnol 88:105–115 Kelley LA, Mezulis S, Yates CM, Wass MN, Sternberg MJE (2015) The Phyre2 web portal for protein modelling, prediction and analysis. Nat Protoc 10(6):845–858 Kleywegt GJ, Jones TA (1996) Phi/psi-chology: Ramachandran revisited. Structure 4:1395–1400
In-Silico Construction of Hybrid ORF Protein …
89
Liu L, Shi X, Bharadwaj KG, Ikeda S, Yamashita H, Yagyu H, Schaffer JE, Yu YH, Goldberg IJ (2009) DGAT1 expression increases heart triglyceride content but ameliorates lipotoxicity. J Biol Chem 284:36312–36323 Lovell SC, Davis IW, Arendall WB III, de Bakker PIW, Word JM, Prisant MG, Richardson JS, Richardson DC (2003) Structure validation by Cα geometry: φ, ψ and Cβ deviation. Proteins 50:437–450 Miao X, Wu Q (2004) High yield bio-oil production from fast pyrolysis by metabolic controlling of Chlorella protothecoides. J Biotechnol 110:85–93 Radakovits R, Jinkerson RE, Darzins A, Posewitz MC (2010) Genetic engineering of algae for enhanced biofuel production. Eukaryot Cell 9:486–501 Ramachandran GN, Ramakrishnan C, Sasisekharan V (1963) Stereochemistry of polypeptide chain configurations. J Mol Biol 7:95–99 Sanda S, Leustek T, Theisen MJ, Garavito RM, Benning C (2001) Recombinant Arabidopsis SQD1 converts UDP-glucose and sulfite to the sulfolipid head group precursor UDP-sulfoquinovose in vitro. J Biol Chem 276:3941–3946 Schrödinger L (2011) Schrodinger software suite. Schrödinger, LLC, New York Schrödinger P (2014) Version 3.5. LLC, New York, NY Smith SJ, Cases S, Jensen DR, Chen HC, Sande E, Tow B, Sanan DA, Raber J, Eckel RH, Farese RV Jr (2000) Obesity resistance and multiple mechanisms of triglyceride synthesis in mice lacking DGAT. Nat Genet 25:87–90 Stone SJ, Myers HM, Watkins SM, Brown BE, Feingold KR, Elias PM, Farese RV Jr (2004) Lipopenia and skin barrier abnormalities in DGAT2-deficient mice. J Biol Chem 279:11767– 11776 SWISS-MODEL | Course | Secondary Structure and Backbone Conformation (2018) https://swi ssmodel.expasy.org/course/text/chapter1.htm. Accessed 14 Dec 2018 US Patent (2018) https://patents.google.com/patent/US9771605. Accessed 14 Dec 2018 Wang Y, Xu HY, Zhu Q (2007) Progress in the study on mammalian diacylgycerol acyltransgerase (DGAT) gene and its biological function. Yi Chuan 29(10):72–1167 Wang Y, Xiao J, Suzek TO, Zhang J, Wang J, Bryant SH (2009) PubChem: a public information system for analyzing bioactivities of small molecules. Nucleic Acids Res 37(2):623–633 Xu J, Francis T, Mietkiewska E, Giblin EM, Barton DL, Zhang Y, Zhang M, Taylor DC (2008) Cloning and characterization of an acyl-CoA-dependent diacylglycerol acyltransferase 1 (DGAT1) gene from Tropaeolum majus, and a study of the functional motifs of the DGAT protein using site-directed mutagenesis to modify enzyme activity and oil content. Plant Biotechnol J 6:799–818 Yu B, Xu C, Benning C (2002) Arabidopsis disrupted in SQD2 encoding sulfolipid synthase is impaired in phosphate-limited growth. Proc Natl Acad Sci 99(8):7–5732 Zou J, Wei Y, Jako C, Kumar A, Selvaraj G, Taylor DC (1999) The Arabidopsis thaliana TAG1 mutant has a mutation in a diacylglycerol acyltransferase gene. Plant J 19:645–653
ANFIS Detects the Changes in Stressful Patterns of Sleep EEG Prabhat Kumar Upadhyay and Chetna Nagpal
Abstract An automated analysis and detection of sleep electroencephalogram (EEG) and stress levels have been performed in this work on pre-recorded EEG data. Based on physiological indicators and powers of sub-band frequencies computed through wavelet transform, features were extracted. With the help of these features, physiological changes in the subjects have been investigated to frame the fuzzy rules to differentiate chronic and acute stress from their respective control groups. Identification of sleep stage is followed by stress level detection using fuzzy logic. The proposed system uses Mamdani fuzzy model and adaptive neuro-fuzzy inference system (ANFIS), that achieves an accuracy of 89.7% for acute stress and 88.0% for chronic stress as compared to their control groups. Keywords Sleep EEG · Stress · Wavelet transform · Fuzzy logic · ANFIS · Mamdani fuzzy model
1 Introduction Study on the effect of heat stress on human nervous system has become an essential component to understand the cause of many psychiatric problems which arise due to the hot environment as one of the natural stress markers. It has been observed that animals respond by activating several processes pertaining to neurophysiology when exposed to hot environments. The influencing factors such as intensity, duration, adaptations to hot environment, play the key role in metabolic change, and obtain the level of thermoregulatory activity, which impacts the functioning of all animals (Menon and Dandiya 1969). Sleep plays an important role in ensuring good physical health. An average adult utilizes 7–8 h per day sleeping during which the sleep cycle, P. K. Upadhyay Department of EEE, Birla Institute of Technology, Mesra, Ranchi, India e-mail: [email protected] C. Nagpal (B) Department of EEE, Birla Institute of Technology, Offshore Campus, Ras Al Khaimah, UAE e-mail: [email protected] © Springer Nature Singapore Pte Ltd. 2021 A. A. Rizvanov et al. (eds.), Advances in Biomedical Engineering and Technology, Lecture Notes in Bioengineering, https://doi.org/10.1007/978-981-15-6329-4_9
91
92
P. K. Upadhyay and C. Nagpal
which consists of various stages of sleep, repeats four to six times. Due to heat stress, alterations in sleep cycles take place. Some of these changes have been noticed to be permanent and some are transitory. Sleep-related disorders can affect the emotional and mental state. Reflection of these sleep disorders appears in the form of deviations in normal sleep cycles (Menon and Dandiya 1969). A normal sleep stage consists of five stages: NREM-1 (Non rapid eye movement1), NREM-2, NREM-3, NREM-4, and REM sleep. NREM-1 or Awake is the stage between wakefulness and drowsiness, NREM-2, 3, and 4 or Slow wave sleep (SWS) is characterized by decrease in responsiveness to environment, REM (Rapid eye movement is characterized by rapid eye movement and also referred to as dreaming phase. The American Association of Sleep Medicine (AASM) characterizes the different stages of sleep based on different physiological changes during these stages (GriggDamberger 2012). Due to the computational ability and learning mechanism of artificial neural networks and fuzzy logic in detecting the sleep patterns, many researchers have been using these soft-computing tools in automated analysis of EEG. Neural network with supervised and unsupervised learning mechanisms have been applied while analysing sleep EEG of animal model and detecting heat stress (Subasi et al. 2005). On the same animal model, EEG power spectrum was also classified which differentiates the depressed subjects from control groups (Upadhyay et al. 2010a). Literature reveals that very few works have been carried out that differentiates normal candidates from heat-stressed conditions with the help of soft-computing techniques. Electroencephalogram (EEG) responses are non-stationary signals which are generated randomly in a response to stimulus (Morstyn et al. 1983). Electromyogram (EMG) and Electrooculogram (EOG) activities are also recorded as polysomnographic parameters to study the sleep patterns. Previous research works suggest that frequency spectrum of EEG covers a wide range of frequencies (Patil and Patil 1998; Tagluk et al. 2009). These frequency bands have been named as delta (0.5–4 Hz), theta (4–8 Hz), alpha (8–12 Hz), and beta (12–40 Hz). In order to train the proposed adaptive system in the classification of sleep stages, a set of feature vectors from sub-band frequencies are extracted from the recorded polysomnograms with the help of which Awake, Slow wave sleep (SWS) and REM were classified (Aboalayon et al. 2016). Using such classification approach, diagnosis of sleep-based disorders becomes quite easy and reliable for clinicians and medical professionals. Many researchers have carried their studies in the field of sleep EEG where artificial neural networks along with other statistical methods have been extensively applied (Aboalayon et al. 2016). However, the advent of fuzzy system which is a rule-based technology has made the analysis of such problems even simpler. Several reports have been published in past few years which investigate how the brain electrical activities get affected when heat is induced externally. The study correlates the changes in EEG pattern as a function of ambient temperature. Due to an increase in the ambient environmental heat, body temperature increases either artificially or spontaneously and as a result of which, an increase in EEG frequency has been observed. Acute heat stress is developed when the subject is kept under stress for a smaller period whereas if stress is applied for a long time, chronic stress is developed in the subject (Upadhyay et al. 2010b).
ANFIS Detects the Changes in Stressful Patterns of Sleep EEG
93
The proposed work involves soft-computing techniques such as fuzzy logic and neural networks to classify the sleep patterns and types of stress induced due to exposure of high heat. The aim of this study is to develop an automated medical investigative system to detect the anomalies in neuronal functioning under the effect of high environmental heat conditions. Alterations in powers of different frequency sub-bands were evaluated with respect to the normal subjects. Further, fuzzy rules which have been framed in terms of variation in power were used to predict the stressed spectra as acute or chronic.
2 Materials and Methods 2.1 Features Percentage change in band power as well as relative band power has been computed for each frequency band, which acts as important features. Relative band power indicates what amount of any frequency sub-band of sleep EEG constitutes in the entire frequency band. It is obtained when absolute power of a particular band is divided by the integrated power computed for the whole spectrum under consideration. Quantitative presence of frequency components in sleep EEG helps in creating fuzzy sets for adaptive fuzzy system (Yang et al. 2002). This formulates the base of determining if a specific frequency constitutes a small or large share of the frequency band. It has been observed that frequency sub-bands—alpha and beta cover a major part of awake state, delta is largely found in slow wave sleep, and presence of theta is found to be very low in awake state. Apart from the features extracted from EEG data set, some statistical features were also extracted from the other two channels i.e., EOG (Electrooculogram) and EMG (Electromyogram) because in order to predict sleep stages, recordings of EOG and EMG activities along with EEG play a vital role to identify the sleep events (Tagluk et al. 2009; Jansen 1985). Mean of the absolute values of EOG and EMG have been calculated for any given sample and considered as one of the features.
2.2 Neuro-fuzzy System The Neuro-Fuzzy implementations on biomedical applications are successful as it combines two approaches i.e., ANN and fuzzy set theory. ANFIS utilizes the computational properties of neural network with fuzzy rule base engine which gives the significant results to model the nonlinear systems. This approach tunes the entire model and minimizes the error between the calculated output and the desired output by learning the input feature vectors through adjusting the weight vectors. Hybrid learning algorithm is used to identify the parameters of Sugeno model (ANFIS)
94
P. K. Upadhyay and C. Nagpal
which is an adaptive system. In order to train the function parameters of FIS and emulate the training data set, the algorithm follows a mixture of back propagation steepest descent method and the least-squares method.
2.3 Rules for Manual Scoring EEG signal consists of a range of frequency bands which are used for the purpose of clinical diagnosis. Manual scoring rules (Tagluk et al. 2009; Estévez et al. 2002; Hobson 1969) were followed to obtain the fuzzy rules to deal with the fuzziness inherently present in the system. In order to classify sleep EEG, following rules has been laid down. Sleep stage will be recognized as: a. Awake: If the powers of alpha and beta sub-bands are high, power of theta sub-band is low and EMG is high. b. SWS: If the power of delta sub-band is high, EMG activity is low. c. REM: If the power of alpha and beta sub-band is high, power of theta sub-band is low and EOG is high.
2.4 Observations on Fuzziness in Stress As a result of externally induced heat stress, subjects undergo physiological changes. These changes may be temporary or permanent, based on the types of heat exposure given to the subject. Several research works have been done in this field where authors have studied the consequence of heat stress in terms of changes in sleep pattern (Upadhyay et al. 2010b; Sinha and Ray 2004). With the help of these changes observed during the experiment, membership functions for input and output fuzzy variables have been designed and the fuzzy rules have been framed to classify stress. When the subjects are under exposure to heat stress for longer period (chronic stress), the changes in sleep patterns were observed to be different than when they were exposed to high environmental heat for shorter duration. These changes in sleep EEG were recorded for chronic and acute situations. In both the cases, variation in powers of delta, theta, alpha, and beta sub-bands were studied to analyze the changes in power with respect to their respective control groups in all frequency bands to study the influence of environmental heat on stress level. It was observed that there has been a noticeable change in powers of these frequency bands relative to their control groups where subjects were not exposed to any heat stress. For awake, SWS and REM stages, variation in powers of these four frequency sub-bands has been tabulated below (Tables 1, 2 and 3).
ANFIS Detects the Changes in Stressful Patterns of Sleep EEG Table 1 Stress classification (fuzzy rules for awake)
Table 2 Stress classification (fuzzy rules for SWS)
Table 3 Stress classification (fuzzy rules for REM)
Chronic
95 Acute
∂ power
↓
No change
θ power
No change
No change
α power
No change
No change
β power
No change
↑
Chronic
Acute
∂ power
↓
↑
θ power
No change
↑
α power
↓
↑
β power
↑
↓
Chronic
Acute
∂ power
No change
No change
θ power
↑
↑
α power
↓
↑
β power
↓
↑
3 Methodology The present study consists of the following four steps which are explained in the respective sections. a. b. c. d. e.
Data acquisition and organization Features extraction Sleep stage classification Stress level classification Result interpretation
3.1 Data Recording This research work uses the same animal model as used by Sinha (2008). Male Charles Foster adult healthy rats of weight between 180–200 g have been selected as subjects in the experiment for data recording under stress and normal conditions. Since neurological development in rats is identical to human being, many researchers have used rats to study their behavioral pattern (Sinha 2003; Kirov and Slavianka 2002). Exposure of heat stress was given to the subjects under the following clinical environment:
96
P. K. Upadhyay and C. Nagpal
a. Control group: The temperature of the BOD incubator in which rats were kept, was maintained at 23 ± 1 °C. b. Acute stress: The temperature of the BOD incubator for a single day was maintained at 38 ± 1 °C for continuous four hours. Data was recorded on the next day. c. Chronic stress: The temperature of the BOD incubator was maintained at 38 ± 1 °C for one hour daily for the period of 21 days. Recording was done on the 22nd day.
3.2 Feature Selection From the continuous recording of EEG signal under stress and normal conditions, the processed data is saved in an interval of 30 s. Features are extracted from these datasets and stored along with their labels. After adjusting baseline and bandpass filtering of the signal, power is calculated for each epoch and power spectrum is analyzed. For each sub-band frequencies, relative band power and percentage change in mean band power were obtained. Therefore, total ten features from each sample dataset were extracted: mean band power delta (pdelta), mean band power of theta (ptheta), mean band power of alpha (palpha), mean band power of beta (pbeta), relative band power of delta (rdelta), relative band power of theta (rtheta), relative band power of alpha (ralpha), relative band power of beta (rbeta), mean of EMG (mEMG), and mean of EOG (mEOG).
3.3 Sleep Stage Classification Classification of sleep stages have been performed efficiently with the help of Sugeno type fuzzy model which uses six inputs as shown in Fig. 1. Feature vectors contain relative powers of delta, theta, alpha, and beta as well as mean of EEG and mean of EMG. A fuzzy object is created and these extracted features have been added to it. Each feature vector is associated with a Gaussian bell membership function. Universe of discourse for these two fuzzy sets representing high and low is obtained by observing the range of the respective input vectors. Since the output which is obtained from the fuzzy system ranges from 0 to 1, range of such outputs as 0–0.33, 0.33–0.66, and 0.66–1 were labeled as Awake, SWS, and REM, respectively. From the feature space, 70% of data is utilized for training the fuzzy system whereas the rest 30% is used for testing. While training, it was observed that the percentage error reduces drastically and after some epochs, the algorithm saturates. It implies that any further training would not make much difference in the weights adaptation of neuro-fuzzy system. In order to extract band power and relative power using wavelet transform, samples length of 30 s has been separately processed. All visually examined samples were used to train and test the performance
ANFIS Detects the Changes in Stressful Patterns of Sleep EEG
97
rdelta rtheta ralpha rbeta mEOG mEMG
Fig. 1 Fuzzy system for sleep stage classification
of the system. The output for a given set of input vectors is shown in Fig. 2 for the sleep stage classification. The change in any one of the inputs in rule base engine reflects the change in output. Each input is associated with the three membership functions as low, medium, and high. An example of this has been illustrated in Fig. 2 where it produces REM as output because the calculated output which is 0.947, falls in the range of REM as per the criteria mentioned above.
Fig. 2 Sleep stage classification using Sugeno model
98
P. K. Upadhyay and C. Nagpal
3.4 Stress Level Classification Stress classification is performed through a Mamdani fuzzy inference system. Having labeled the feature vectors through the steps as mentioned in the previous section, separation of features representing the same sleep stage was carried out for each of the stress spectra. Further, a fuzzy model is built to classify the sleep stages such as Awake, SWS, and REM into an acute and chronic group. Figure 3 shows the fuzzy rule implementation for the stress classification of each stage. This system is presented with ten inputs and one output, where first six inputs are associated with the sleep stage classification and four additional features have been incorporated to classify the stress into chronic or acute group. An average band power was calculated for each control group of respective sleep stage and kept as a reference so as to quantitatively analyze the deviation when the subjects were exposed to stress. The change (increase or decrease) in-band power is reflected by the deviation of band power in acute and chronic cases for respective sleep stage. The percentage change in band powers is further fed to the input of the fuzzy system. In order to represent the percentage change, two fuzzy membership functions with linguistic variables—high and low have been used where the universe of discourse was properly taken from the available input fuzzy sets. Based on the nature of input data set and in view of the inherent advantages of a triangular membership function, function ‘trimf’ was used to express the fuzzy sets low and high. The fuzzy rules mentioned in Tables 1, 2 and 3 were added in the fuzzy system for stress detection and classification of Awake, SWS and REM stages. In
Fig. 3 Fuzzy system used for stress detection
ANFIS Detects the Changes in Stressful Patterns of Sleep EEG
99
Table 4 Stress classification Percentage change in band power
AWAKE Chronic
SWS
REM
Acute
Chronic
Acute
Chronic
Acute
pdelta
L
X
L
H
X
X
ptheta
X
X
X
H
H
H
palpha
X
X
L
H
L
H
pbeta
X
H
H
L
L
H
Percentage Accuracy
Fig. 4 Accuracy breakdown
Sleep Stages
Table 4, percentage change in delta, theta, alpha, and beta band power in comparison to their respective control group have been presented, where ‘L’ shows the decrease in percentage power and ‘H’ shows the increase in percentage power.
4 Results A plot as shown in Fig. 4 shows the performance of stress classification in which percentage accuracy of two stress classes belonging to three sleep stages have been clearly depicted.
5 Discussion Many approaches such as fuzzy logic, neural network (Sukanesh and Harikumar 2007; Yang et al. 2002), SVM (support vector machine) (Koley and Dey 2013), RBF (Radial basis function), etc., have been implemented by researchers for the detection
100
P. K. Upadhyay and C. Nagpal
of sleep stages. The rules for manual scoring which was established by R&K were followed in ANFIS and the results achieved an accuracy of 88.5% [30]. The modified rule which was established by AASM has been followed in this study and has resulted to perform the sleep stage classification with an accuracy of 90.65%. After sleep stages have been classified, stress level classification has been done. Use of wavelet transform as a preprocessor and as an analyzer too has been very common in the analysis of problems such as sleep EEG analysis, spike detection, sleep spindles, epileptic seizures detection, event-related potentials, etc., (Heiss et al. 2002) (Sukhorukova et al. 2010). Several linear and nonlinear classifiers with parametric and non-parametric approaches have been frequently used to classify different EEG events which adequately detect the changes in neuronal functioning as reported in literature (Koley and Dey 2011). However, literature does not witness any work in relation to classification of heat-stressed sleep EEG spectra with the help of fuzzy models such as Sugeno and Mamdani models. Previous works in the same area as reported by a few researchers (Upadhyay et al. 2010a; Sinha and Ray 2004; Sinha 2008) employ multi-layered perceptron neural network with back propagation algorithm and unsupervised learning schemes through which heat-stressed patterns were detected. But so long as the investigation on alteration in sub-band frequency and powers of sleep EEG owing to hot environment using fuzzy logic is concerned, no work has been reported so far. In this study, a very encouraging result has been achieved with this model in which percentage accuracy is obtained as 89.2%. The proposed method enables the system to considerably discriminate the stress patterns from their respective control.
6 Conclusion In this paper, we have successfully classified stressed pattern from normal using neuro-fuzzy concepts. Classification results are found to be in agreement with visual scoring. This concept may further be extended to study the effects of other stress stimuli such as noise, immobility, pollutants, etc., on brain electrophysiology. Application of other sophisticated DSP tools such as s-transform may be applied to investigate the phases of EEG signal, if required for brain signal analysis. With this approach, an online automated diagnostic system can be developed to classify or identify the abnormalities in brain-functions.
ANFIS Detects the Changes in Stressful Patterns of Sleep EEG
101
References Aboalayon KAI et al (2016) Sleep stage classification using EEG signal analysis: a comprehensive survey and new investigation, entropy Estévez P, Held C, Holzmann C, Perez C, Pérez J, Heiss J, Garrido M, Peirano P (2002) Polysomnographic pattern recognition for automated classification of sleep-waking states in infants. Med Biol Eng Comput 40(1):105–113 Grigg-Damberger M (2012) The AASM scoring manual four years later. J Clin Sleep Med Heiss J, Held C, Estevez P, Perez C, Holzmann C, Perez J (2002) Classification of sleep stages in infants: a neuro fuzzy approach. IEEE Eng Med Biol Mag 21(5):147–151 Hobson A (1969) A manual of standardized terminology, techniques and scoring system for sleep stages of human subjects. Electroencephalogr Clin Neurophysiol 26(6):644 Jansen B (1985) Feature extraction methods for EEG analysis. Electroencephalogr Clin Neurophysiol 61(3):S222 Kirov R, Moyanova S (2002) Distinct sleep-wake stages in rats depend differentially on age. Neurosci Lett 322(2):134–136 (Web) Koley B, Dey D (2011) An ensemble system for automatic sleep stage classification using single channel EEG signal. Comput Biol Med 42(12):1186–1195 Koley B, Dey D (2013) Automatic detection of sleep apnea and hypopnea events from single channel measurement of respiration signal employing ensemble binary SVM classifiers. Measurement 46(7):2082–2092 Menon MK, Dandiya PC (1969) Behavioural and brain neurohormonal changes produced by acute heat stress in rats: influence of psychopharmacological agents. Eur J Pharmacol 8:284–291 Morstyn R, Duffy FH, McCarley RW (1983) Altered topography of EEG spectral content in schizophrenia. Electrocephalogr Clin Neurophysiol 56:263–271 Patil SA, Patil SP (1998) Computerised EEG analyser. IETE Tech Rev 15(6):503–507 Sinha RK (2003) Artificial Neural Network detects changes in electro-encephalogram power spectrum of different sleep-wakes in an animal model of heat stress. Med Biol Eng Comput 41:595–600 Sinha RK (2008) EEG power spectrum and neural network based sleep-hypnogram analysis for a model of heat stress. J Clin Monit Comput 22:261–268 Sinha RK, Ray AK (2004) An assessment of changes in open-field and elevated plus-maze behavior following heat stress in rats. Irani Biomed J 8:127–133 Subasi A et al (2005) Automatic recognition of vigilance state by using a wavelet-based artificial neural network. Springer, 25 Jan 2005 Sukanesh R, Harikumar R (2007) Analysis of fuzzy techniques and neural networks (RBF&MLP) in classification of epilepsy risk levels from EEG signals. IETE J Res 53(5):465–474 Sukhorukova S, Ofoghi V, Saleem U, Muecke A, Philippe BM, Huda B, Lévy G (2010) Automatic sleep stage identification: difficulties and possible solution. In: Proceedings of the 4th Australasian workshop on health informatics and knowledge management (HIKM 2010), Brisbane, Australia Tagluk M, Sezgin N, Akin M (2009) Estimation of sleep stages by an artificial neural network employing EEG, EMG and EOG. J Med Syst 34(4):717–725 Upadhyay PK, Sinha RK, Karan BM (2010) Predicting heat-stressed EEG spectra by self-organising feature map and learning vector quantizers SOFM And LVQ based stress prediction. J Biomed Sci Eng 03(05):529–537 Upadhyay PK, Sinha RK, Karan BM (2010) Detection and analysis of the effects of heat stress on EEG using wavelet transform—EEG analysis under heat stress. J Biomed Sci Eng 405–414 Yang E-J, Shin D-S, Kim E-S (2002) The characteristic analysis of EEG artifacts. J Fuzzy Log Intell Syst 12(4):366–372 (Web)
Recent Advances in Deep Learning Techniques and Its Applications: An Overview Abhishek Hazra, Prakash Choudhary, and M. Sheetal Singh
Abstract Learning with images and their classification, segmentation, localization, annotation, and abnormally detection is one of the current challenging and exciting task for the researchers. Recently deep learning techniques give excellent performance in Object Detection, Speech Recognition, Abnormality Detection, Business Analysis, and almost all other domains. But one important implication of deep learning techniques can found in Medical Image Analysis. Deep learning techniques beat the human-level performance and come with a better solution in the medical domain. Among different deep learning techniques Convolutional Neural Network, Recurrent Neural Network, Long Short-Term Memory, Deep Belief Network models are topmost priority for the researchers. In this paper, we briefly examine different application area of deep learning techniques and some current state-of-the-art performances of it. Moreover, we also discuss some of the limitations of Deep Learning techniques. As expected this paper creates a clear understanding of Deep Learning techniques and its applications. Keywords Deep learning · CNN · RNN · Health care
1 Introduction Neuroscience researchers examine that visual representation of brain can be done in two pathways, Dorsal pathway and Ventral pathway. The information of location movement observation follows the dorsal pathway and detection, color, texture shape, A. Hazra Indian Institute of Technology (ISM), Dhanbad 826004, Jharkhand, India e-mail: [email protected] P. Choudhary (B) · M. Sheetal Singh National Institute of Technology Manipur, Imphal 795004, India e-mail: [email protected] M. Sheetal Singh e-mail: [email protected] © Springer Nature Singapore Pte Ltd. 2021 A. A. Rizvanov et al. (eds.), Advances in Biomedical Engineering and Technology, Lecture Notes in Bioengineering, https://doi.org/10.1007/978-981-15-6329-4_10
103
104
A. Hazra et al.
such information follow the ventral pathway. In 1959 Hubel and Wiesel (Hubel and Wiesel 1968) first observe that visual cortex’s in brain mainly responsible for detecting lights. Inspired by this innovation LeCun et al. (Le Cun et al. 1990) in 1990 first make a handwritten character recognition system. This was the first neural network architecture which tried to learn and recognize characters. Since 90s many researchers come with their deep advance architecture which solves many real-time applications. AlexNet (Krizhevsky et al. 2012), VGGNet (Simonyan and Zisserman 2015), ResNet (He et al.2015c), models break all the previous performance in various domains. Recently Deep Learning Techniques come with a better solution for analyzing different kinds of data. The idea of deep learning has a very old history. Because of its high computation power and colossal amount of data, deep learning techniques were not so popularly used back then. But in late 20s deep learning techniques accelerate its performance with the help of Graphical Processing Unit and massive amount of data. Deep Learning techniques gives state-of-the-art performance in almost all the domains like Object Detection, Speech Recognition, Fraud Detection, Face Recognition, Sentiment Analysis. Currently, deep learning techniques give excellent performance in medical image analysis. From 2015, research in deep learning for medical domain has increases exponentially. The number of research papers, journals, and articles in this domain are increasing day by day. It means this field is gaining interests gradually at a very rapid rate. There are special issues in almost all the journals with deep learning as a keyword. These are the primary motivation for us to work in this particular field. In this paper, we briefly examine different application area of deep learning techniques. Our search list contains “Deep Learning” either in the title or the keyword in various articles, journals good conference proceedings which are mainly focused on this field only. In this review, we try to scrutinize all the popular related papers published till 30 March 2018. We expect that this paper makes a brief overview of deep learning techniques in all the domains. Our main aim for this review: • To show deep learning techniques and its performance. • To display current research Scenarios. • To highlight some of the current research challenges. The rest of the paper structured as follows. In Sect. 2 we introduce some essential deep learning techniques. Section 3 describes some popular application area of deep learning techniques. Section 4 narrates the deep learning implementation tools. Section 5 shows an essential overview discussion in medical imaging. Finally, conclusion and feature work were speculated in Sect. 6.
Recent Advances in Deep Learning Techniques …
105
Fig. 1 Famous LeNet-5 network architecture with intermediate visual representation
2 Overview of Deep Learning Techniques In recent times deep learning techniques are widely used in all the domains like Object Detection (Yoo et al. 2015), Sentiment Analysis (Wang et al. 2016c), Medical Image Analysis (Lo et al. 1995), Speech Recognition (Waibel et al. 1990), SelfDriving car, Automatic Machine Translation, Automatic Text Generation (He et al. 2015b), advertising, and many more. With the advancement of GPU based systems, several deep learning techniques are also introduced to address different kinds of problems. Convolutional Neural Network, Recurrent Neural Network, Restricted Boltzmann Machine, LSTM, Deep Autoencoder networks are current state of the learning algorithm. In this section, our main aim is to address some of the most popular deep learning techniques which create a huge impact on current research.
2.1 Convolutional Neural Network Convolutional Neural network (CNN) (Vincent et al. 2008) is one of the most popular learning algorithms in computer vision field. Currently, many researchers come with their individual CNN architecture though there is a similarity between all the networks. Basically convolutional neural network consists of four types of layers. Convolutional layer, Activation layer, Pooling layer, and Fully Connected layer. LeNet-5 as presented in Fig. 1, was the first generalizes neural network architecture (Le Cun et al. 1990) which is still popularly used in current times. Convolutional layer takes information from the input data and produces a feature map with the help of kernels. The number of convolutional layers is vary from architecture to architecture. Generally first level of convolutional layers learns low-level features like dark and bright pixels, second layer of convolutional layers may learn horizontal edges, vertical edges, next level of convolutional layer learns some more complex functions like ears, nose mouth. As the number of layers increased neural network learns even more complex functions like face, object, and characters. These feature maps are passed through a nonlinear activation function which gives acceleration to the CNN to understand complex functions. Finally one or more number of fully connected layers which summarize the learnable information and put into a softmax classifier. The softmax classifier gives the output probability of each and every class for the given input.
106
A. Hazra et al.
2.2 Recurrent Neural Network Recurrent neural network (RNN) (Poultney 2006) is one of the most interesting neural network architecture. RNN is interesting because of its use in many applications and is also notable performance under challenging applications. RNN is mostly trained by a sequence of data like sentence and make subsequent similar sentences which are most likely used in chatbots. RNN is widely used in several applications like image captioning, generating review, generating feedback, generating Music. Feedforward neural networks are not designed for sequence/time-series data, hence results with time series/sequence data are inadequate and moreover, they cannot design for storing memory. To address this problem, recurrent neural network was designed. Recurrent neural networks are the type of networks designed for capturing information from sequence/time series data. In RNN sequences is feed as current input, calculate a hidden state, and compute the output. For the next time step, the network takes new information as well as information from the previously hidden state to compute the current hidden state to predict the next output. Finally, a loss functions to improve the accuracy of RNN. This types of networks use in time series prediction like weather forecasting, Stock prediction, and sequence of data generating application like music, video.
2.3 Long Short-Term Memory Long short-term memory (LSTM) is a type of recurrent neural network. LSTM is the next logic step in the progression of neural network Learning. It is technique of learning sequence of data or video frame and capable of learning long-term dependencies. One interesting idea of weighted self-loop to introduce path where Gradient flow for a long time in Long short-term memory. By addition of self-loops current hidden layers are controlled by previously computed hidden layers. Even for fixed parameters, the time scale of integration can be changed according to the input time is the output of this model. Different researchers found that LSTM networks are incredibly successful in many applications such as speech recognition, music generation, machine translation, image captioning, handwritten recognition. There are several deep learning techniques like Deep Autoencoder (Salakhutdinov and Hinton 2009; Rifai et al. 2011; Masci et al. 2011; Chen et al. 2013), Boltzmann Machine (Younes 1999; Center Berkeley 2016), and Deep belief networks are also popularly used in various domains.
Recent Advances in Deep Learning Techniques …
107
3 Applications of Deep Learning Techniques Currently, deep learning techniques are giving an excellent performance in Action Recognition, Significant Data Analysis, Sentiment Analysis, Medical Image Analysis, Character Recognition, Image Classification, Object Detection, Object Tracking, Pose Estimation, Visual Salient Detection Sense leveling, Speech Recognition, Natural Language Processing, Remote Sensing. In this segment, we try to introduce some of the current research application of deep learning techniques as illustrated in Fig. 2.
3.1 Action and Gesture Recognition One of the interesting application of Deep learning techniques is in action recognition. Most of the Companies are using some action recognition for their internal security purpose. Because of its high demand and current challenges attract deep learning researchers in this field. This field has been examining from last decades and reported huge progress within the computer vision field. RNN (Le Cun et al. 1990), LSTM (Rowley et al. 1998), 3D convolutional neural network (Yang et al. 2017a), pertained features are the topmost priorities of deep learning researchers. Mainly three types of network model were used for action and texture recognition. 3D convolutional layer (de Brebisson and Montana 2015; Gao et al. 2015; Lo et al. 1995; Chen et al. 2017; Gao 2016; Tarando et al. 2016; Zhu et al. 2017; Xu et al. 2016b; Dittrich et al. 2011), motion-based input feature (Alexe et al. 2012; Zhao et al. 2016; Xu et al. 2016b; Chen et al. 2017; Hinton and Salakhutdinov 2006) and temporal methods which is the combination of 2d or 3d CNN networks. Though RNN is one of the important deep learning architecture particularly used for this task, this kind of network suffers from short-term memory loss. To address this problem LSTM (Anavi et al. 2015) was introduced. LSTM works in the inner layer of RNN. B-RNN (Goodfellow et al. 2014), H-RNN (He et al. 2015a), D-RNN (Janowczyk et al. 2017) are some extended, modified version of LSTM. Moreover fusion-based deep learning techniques (Lo et al. 1995; Pluim et al. 2003; Ngo et al. 2017; Yan et al. 2014; Chen et al. 2017; Center Berkeley 2016) are also popularly used for action recognition.
3.2 Deep Learning for Big Data Big Data refers considerable amount of datasets (Philip Chen and Zhang 2014) which can synthesize specific patterns. Deep learning techniques are widely used for analyzing big data and succeeded to find certain hidden pattern that was impossible so far. Proper knowledge plays a critical role for success in many companies as well. This need can be satisfied by combining this two domain: Deep Learning and Big Data.
108
A. Hazra et al.
Big companies like Facebook, Google, and Yahoo used deep learning techniques and getting benefited from it. The analysis of big data can be subdivided into three phase: Big Data processing, Big Data storage, Big Data management. For better decision making we need large and good quality of data which requires data preprocessing (Hinton and Salakhutdinov 2006; Witten et al. 2016; Riabov and Liu 2006; Han et al. 2014; Siddiqa et al. 2016; Michael and Miller 2013; McAfee et al. 2012). Data cleansing, transmission sequencing are some of the intermediate steps of data processing. Storing big data in PT scale is not a feasible solution for researchers and interesting communities. Though recent advances of cloud computing anyhow reduce some problem. The main interesting thing in it is to create a storage management system which provides enough data and utilizes information retrieval (Dittrich et al. 2011), replication, indexing are the intermediate steps of storing big data (Li et al. 2008; Deng et al. 2014; Chen 2010) processing is one of the challenging tasks. There are several processing issues in managing big data. Recently AI companies invested a huge amount of money in big data processing (Buza et al. 2014; Porkar 2012; Jafari et al. 2016; Waibel et al. 1990). For addressing such problem, many machine learning (ML) researchers come with their handcraft feature learning techniques but fails to give a good result in practical aspects. But deep learning techniques give a better solution for handling both labeled and unlabeled datasets.
3.3 Deep Learning for Sentiment Analysis Deep Learning Techniques (Morin and Bengio 2005; Mikolov et al. 2013a, b; Mnih and Kavukcuoglu 2013; Moraes et al. 2013; Johnson and Zhang 2015) are also useful for analyzing emotions. Though understanding of human emotion and explain it in terms of words is a challenging task for the computer vision researchers. Sometimes words are not enough to correctly explain our emotions as some emotions has no language translation. But deep learning techniques assistance to understand human emotional data which helps to take optimal decisions. There are mainly two basic approaches of sentiment analysis. Lexicon-based approach and AI-based Approach. In lexicon-based approach for given sentence words are split into small tokens also knows as tokenization. Bag of words is the count the number of frequencies of each word. Based on this it decides positive and negative sentences. AI-based deep learning techniques are the current trend research in sentiment analysis. For a large dataset deep learning techniques were trained and also be applied for real-time applications. CNN (Kalchbrenner et al. 2014; Kim 2014; dos Santos and Gatti 2014; Wang et al. 2016b, c; Guggilla et al. 2016; Mishra et al. 2017; Bengio et al. 2013; Qian et al. 2015), RNN (Tang et al. 2015; Guan et al. 2016; Yu and Jiang 2016) LSTM (Tang et al. 2016; Salakhutdinov and Hinton 2009; Qian et al. 2017; Li et al. 2017; Wang et al. 2015c; Huang et al. 2017; Le and Mikolov 2014; Glorot et al. 2011) models are popularly used in this task. Though deep learning-based sentiment analysis is a hard process of computation but these techniques give better result than traditional techniques. Document-level Sentiment classification (Wang et al. 2015c; Williams
Recent Advances in Deep Learning Techniques …
109
and Zipser 1989; Liu and Zhang 2017; Masci et al. 2011), Sentence level sentiment classification (Loshchilov and Hutter 2016; Wang et al. 2016c), aspect level sentiment classification (Liu and Zhang 2017; Yang et al. 2017b) are some of the intermediate steps of sentiment analysis. Large social media companies like Facebook twitter google has deep learning-based approaches for analyzing customers perspective.
3.4 Deep Learning for Medical Image Analysis Analysis of medical images and their classification localization segmentation annotation abnormally detection are one of the current research interest. Since 2014 after the development of GPU based systems deep learning techniques give excellent performance in medical domain. Many researchers collect their data and make available for research purpose. Different research shows that CNN (Suk and Shen 2016; de Brebisson and Montana 2015; Choi and Jin 2016; Zhang et al. 2015a; Birenbaum and Greenspan 2016; Brosch et al. 2016) based deep learning models are most widely used in medical engineering. Apart from CNN, RBM, RNN (Stollenga et al. 2015; Andermatt et al. 2016) Autoencoder based models are also popularly different health care applications like brain image analysis(Sarraf and Tofighi 2016; Chen et al. 2016; Ghafoorian et al. 2016a, b), retinal image analysis (Gulshan et al. 2016; Zilly et al. 2017; Chen et al. 2015; Abràmoff et al. 2016; Lu et al. 2016a; van Grinsven et al. 2016; Gulshan et al. 2016; Gao et al. 2015), chest x-ray image analysis (Lo et al. 1995; Anavi et al. 2015; Anavi et al. 2016; Lin et al. 2014; Vaillant et al. 1994; Hwang et al. 2016; Kim and Hwang 2016; Rajkumar et al. 2017; Yang et al. 2017a), CT chest x-ray image analysis (Wang et al. 2017; Charbonnier et al. 2017; Shen et al. 2015a; Chen et al. 2017; Dou et al. 2017; Setio et al. 2016; Sun et al. 2016; Anthimopoulos et al. 2016; Christodoulidis et al. 2017; Gao 2016; Tarando et al. 2016; van Tulder and de Bruijne 2016; Avendi et al. 2016), pathology image analysis (Xie et al. 2016; Wang et al. 2016e; Xu et al. 2016a, b; Chang et al. 2017; Çiçek et al. 2016; Chen et al. 2017; Janowczyk et al. 2017; Hubel and Wiesel 1968), cardiac image analysis (Emad et al. 2015; Ngo et al. 2017; Poudel et al. 2016; Tran 2016; Prasoon et al. 2013), abdominal image analysis (Li et al. 2015; Vivanti et al. 2015; Wang and Gupta 2015; Yu et al. 2017; Zhu et al. 2017; Zhao et al. 2016), musculoskeletal image analysis (Shen et al. 2015b; Suzani et al. 2015; Antony et al. 2016). Figure 2 shows the pictorial application area of deep learning techniques in medical science.
3.5 Deep Learning for Text Detection and Recognition Character and text recognition is one of the current time research and had been studied in the computer vision field from long time. Optical character recognition also popularly known as OCR recognition is one of the fundamental academic research. CNN is the main building block architecture for recognition of characters. We also divide this
110
A. Hazra et al.
Fig. 2 Application area of deep learning techniques in health care sector (Masci et al. 2011)
task into three subcategories: Text detection, text recognition from small region, and combination of text detection and recognition. CNN models are widely used for text detection (Zhang et al. 2015b). There are several standard handwritten and optical character available in almost all the languages in worldwide. One improvement of this work is the combination of CNN and Maximally Stable External Regions (MSER) (Goodfellow et al. 2014; Zhang et al. 2015b), Bag-of-Words (BoG) and CNN based sliding Non-Maximal suppression (NMS) (He et al. 2015b) based CNN structures are also popularly used. Similar to text detection, Text recognition is also popular research area Good fellow et al. (Shi et al. 2015) proposed a multilevel CNN classifier for character recognition from multidigit input string. Conditional Random Fields (CRF) based CNN, feature extraction based CNN (Gers et al. 2000), Sliding window-based LSTM (Jaderberg et al. 2014), and feature extraction based text recognition (Jaderberg et al. 2015) are also some popular techniques of text recognition. End-to-End text spotting with bounding box (Lawrence et al. 1997; Simard et al. 2003) is also the popular research interest in computer vision field. According to Ethnologies catalog of world languages, there is 6909 number of registered script language exist and most of the counties have their own official languages. Thus character recognition field has its own separate interest. Currently, automatic character recognition is used in machine translation, postal systems, identification recognition, image translations.
Recent Advances in Deep Learning Techniques …
111
3.6 Deep Learning for Image Classification Deep Learning Techniques gives a tremendous performance for classification of the object from a large dataset. Though CNN was used for image classification long back (He et al. 2015b), but it creates a remarkable performance in recent times with the advances of Graphical processing units and a large amount of data (He et al. 2015b; Lawrence et al. 1997; Everingham et al. 2014; Deng et al. 2009). In 2012 AlexNet creates a huge impact for mage classification of large-scale images which also wins the ILSVRC 2012 challenge. Taking this motivation of the work many researchers take their interest in increasing the classification accuracy by tuning the hyper parameters in the neural network. Several researchers come up with their new classification technique which sometimes works well. Hierarchy based image classification is a common technique for classifying a large class of images (Wang et al. 2015a). Hierarchy of CNN in discriminate feature learning for sharing their hierarchy of information to share among the classes (Xiao et al. 2014). Fine-gained feature learning (Yan et al. 2014), trained hierarchal network (Nilsback and Zisserman 2008), embedding CNN into a subcategory of the hierarchy methods are also popularly used to reducing the classification error. Subcategory image classification datasets (Yu and Grauman 2014; Yang et al. 2015) also takes lots of interest. CNN (Uijlings et al. 2013), RCNN (Pluim et al. 2003; Lin et al. 2015b) Deep LAC (Lin et al. 2015a) based object part classification is also popularly used. Create a subnetwork, localize a region, and estimate the predictive class (Krause et al. 2015) also helps to improve classification accuracy. Both supervised and unsupervised learning techniques on annotated data are popular to fine-tuned (Dalal and Triggs 2005) the class. Ensemble the localization (Rowley et al. 1998), co-segmentation (Rowley et al. 1998), leveling by simplicity, visual attention based CNN models are also popularly used for image classification.
3.7 Deep Learning for Object Detection One of the current time computer vision problems is object detection. There are many research issues for detecting objects from video or images. Though CNN based object detection techniques started in early 90s. However, due to lack of computational power and a small amount of data breaks the progress of CNN-based system. Recently in 2012 after the huge success in ImageNet challenge (Deng et al. 2009) this field gets back interested from the research community. In earlier times CNN based object detection (Lin et al. 2014; Vaillant et al. 1994) using sliding window were so popular, but these techniques require high computational power which makes them unreliable for massive datasets. Like VOC (He et al. 2015b), IMAGENET (Lawrence et al. 1997), MSCOCO (Alexe et al. 2012). To address this issue, Object proposal based technique introduced. Different literature (Carreira and Sminchisescu 2012; Pluim et al. 2003) shows that object proposal based techniques are the most generic measure of the test, a generic window is used to propose whether an object
112
A. Hazra et al.
present or not., then passes it to next level of generic detection to understand objects are belonging to the same class or not. Region-based CNN (R-CNN) (Sermanet et al. 2013) is one of the popular objection technique. A pertained CNN is used on a selective search to extract feature and SVM used to classify objects. Several improvements were done to improve the performance. Feature extraction (He et al. 2015a; Carreira and Sminchisescu 2012; Sermanet et al. 2013), SPP net (Yoo et al. 2015), pyramidal R-CNN (Felzenszwalb et al. 2010), bounding box (Redmon et al. 2016), bootstrapping (Liu et al. 2015), Yolo, SDD, top-down search (Gidaris and Komodakis 2015) methods are introduced for better performance in dynamically challenging environments (Liu et al. 2015; Loshchilov and Hutter 2016; Lu et al. 2016b).
3.8 Deep Learning for Object Tracking Another success of deep learning techniques can be found in object tracking. CNN and RNN based models are popular in this particular task. CNN based object tracking, target specific (Li et al. 2014) object tracking, temporal adaptation mechanism (Li et al. 2014), tracking based (Plis et al. 2014), similarity-based visual tracking (Hong et al. 2015), are most popular. In almost all the small/big companies and institutes use some kinds of tracking system for detecting persons, counting vehicles, finding missing elements, video surveillance.
4 Software and Implementation Tools Table 1 shows some of the deep learning implementation tools. Keras, Tensorflow, Theano, PyTorch tools are widely used for implementation of AI techniques. Most of the tools use python as their underlying framework. The number of libraries for supporting the python is increased with the acceleration of GPU based systems. One of the main reasons for development of deep learning techniques is the Nvidia Corporation. Almost all the researchers use GPU based systems for accelerating their training time. In Table 1 we try to introduce the interdependencies of different learning techniques.
5 Discussion Overview Our research query also exploits one common problem “what are the best possible ways of training a neural network.” To find this answer to the question, we examine some breakthrough performances and general intuitions for understanding a neural network. We found out that there are mainly two ways we can train our neural network: first is to create own neural network architecture and the second is by using
Recent Advances in Deep Learning Techniques …
113
Table 1 Some of the popular deep learning implementation tools Tools
Platform
Support
Interface
Caffe (Williams and Zipser 1989)
Windows, Linux, Mac OSX
CNN, RNN
Python, C++, Matlab, Cuda
Tensorflow (Salakhutdinov and Hinton 2009)
Windows, Linux, Mac OSX, Android
Almost support all deep learning techniques
Python
Theano (Younes 1999)
Windows, Linux, Mac OSX
Almost support all deep learning techniques
Python, Cuda
Torch (Microsoft 2016)
Windows, Linux, Mac OSX
Almost support all deep learning techniques
Lua
Keras (Delakis and Garcia 2008)
Windows, Linux, Mac OSX
Almost support all deep learning techniques
Cross-platform, Cuda
PyTorch (Xu and Su 2015)
Linux, Mac OSX
Almost support all deep learning techniques
Python, C, Cuda
transfer learning. In the previous section, we already examine some unique deep learning architecture. Now in this segment, we will try to understand the transfer learning techniques. There are mainly three ways of using pre-trained model and train neural network: Fixed feature extractor, Fine-tuning the model, and pertained the model(Choudhary and Hazra 2019). Fixed feature extraction: It is one of the early ML algorithms. First, use a technique which summarize the features and then apply on a classifier for predicting output levels. Also, the same way we can train neural network at first choose a convolutional neural network trained on big dataset like ImageNet and by removing last fully connected layer the network can be treated as a fixed feature extractor. Once the feature was extracted then the neural network trains on a classifier for new dataset. Fine-Tuning the ConvNet: Another important strategy in deep learning is not only retrained as the classifier over new dataset but also to replace and fine-tune the learning experiences of the neural network. It may also be possible to train all the layers or keep some of the earlier layer fixed and fine-tune the upper layers. One notable thing to mention, the earlier layer of convolutional layer contain more generic low-level information’s which can be advantageous for new dataset. Different experiments show that layer-wise fine-tuning of a ConvNet for a big data performs better than making a neural network from the sketch. There are certain intuitions when and how to fine-tune a network, deciding choosing a perfect transfer learning technique on a new dataset is a bit challenging task. There are several strategies one should take care, but the important two are the size of the new dataset and similarity between old and new datasets. In the lower level of ConvNet contain a lower level of
114
A. Hazra et al.
generic information and upper level of the network contain more specific information related to the dataset. Some thumb rule for fine-tuning the new dataset are: • If the dataset is small in size, then fine-tuning a ConvNet over a small dataset will not be a good idea as the deep neural network may suffer from overfitting problem. Hence, using a linear classifier on a small dataset might be a good idea. • If the dataset is large and there is a similarity between two datasets then using a pre-trained model will give more confidence not to be overfitting the network, hence chances of increasing the performance of the network. • If the new data is small and differ from original data, then using a linear classifier may not always work, instead use of support vector machine classifier may be beneficial for new dataset as the network contain data specific information. • If the data is large and differs from original data, then fine-tuning a residual neural network sometimes helpful because it is found out that exploring vanishing gradient can lead some problem for weight updation. Even though making a neural network from scratch also works depending on the dataset. Pretrain Models: Training a neural network on a large image dataset like ImageNet may take ~2–3 weeks for training on a Search Results Web results Graphics processing unit (GPU) based systems. Researchers sometimes release their final work for helping others. Using a pretrain model a fuse of different deep neural network sometimes also beneficial for Training a neural network.
6 Conclusion and Future Work All the challenging issues discussed in the previous section were not been tackled yet by the researchers. Though some successes were achieved by using deep learning techniques. From this survey one observation we can make, many researchers’ uses pretrained networks for evaluate their model. ResNet, VGGNet, GoogleNet networks are the top listed architectures for the researchers. Even though it is not clear that these models will work in all the domains. Recently some good results were achieved by making a fused model of different networks. Though there are some limitations of deep learning techniques, still it is widely used for solving real-time problems. Convolutional neural networks, RNN, LSTM networks create a benchmark performance in computer vision, robotics, speech recognition, and all the domains. In this literature, we also try to introduce the capabilities of different deep learning techniques. As this research field is new, there is a massive gap for improvement. In conclusion, we can say deep learning techniques are the current state of the learning algorithms. We expect that, in feature by using deep learning techniques researchers can solve many unsolved problems. Our work still in progress, in recent features we are trying to detect different chest diseases by using deep learning techniques.
Recent Advances in Deep Learning Techniques …
115
Acknowledgements The author would like to thank unacquainted reviewers for their valuable comments. Author would also like to thank National Institute of Technology, Manipur and also department of Computer Science and Engineering for providing Lab and required infrastructure.
References Abràmoff MD, Lou Y, Erginay A, Clarida W, Amelon R, Folk JC, Niemeijer M (2016) Improved automated detection of diabetic retinopathy on a publicly available dataset through integration of deep learning. Invest Ophthalmol Vis Sci 57(13):5200–5206. https://doi.org/10.1167/iovs.1619964 Alexe B, Deselaers T, Ferrari V (2012) Measuring the objectness of image windows. In: PAMI Anavi Y, Kogan I, Gelbart E, Geva O, Greenspan H (2015) A comparative study for chest radiograph image retrieval using binary texture and deep learning classification. In: 2015 37th annual international conference of the IEEE engineering in medicine and biology society (EMBC), Milan, pp 2940–2943. https://doi.org/10.1109/EMBC.2015.7319008 Anavi Y, Kogan I, Gelbart E, Geva O, Greenspan H (2016) Visualizing and enhancing a deep learning framework using patients age and gender for chest x-ray image retrieval. In: Proceedings of the SPIE 9785, medical imaging 2016: computer-aided diagnosis, vol 978510, 7 July 2016. https:// doi.org/10.1117/12.2217587 Andermatt S, Pezold S, Cattin P (2016) Multi-dimensional gated recurrent units for the segmentation of biomedical 3D-data. In: Proceedings of the deep learning in medical image analysis (DLMIA). Lecture Notes in Computer Science, vol 10008, pp 142–151 Anthimopoulos M, Christodoulidis S, Ebner L, Christe A, Mougiakakou S (2016) Lung pattern classification for interstitial lung diseases using a deep convolutional neural network. IEEE Trans Med Imaging 35(5):1207–1216. https://doi.org/10.1109/TMI.2016.2535865 Antony J, McGuinness K, Connor NEO, Moran K (2016) Quantifying radiographic knee osteoarthritis severity using deep convolutional neural networks. arXiv:1609.02469 Avendi M, Kheradvar A, Jafarkhani H (2016) A combined deep-learning and deformable-model approach to fully automatic segmentation of the left ventricle in cardiac MRI. Med Image Anal 30:108–119 Bengio Y, Courville A, Vincent P (2013) Representation learning: a review and new perspectives. IEEE Trans Pattern Anal Mach Intell (2013) Birenbaum A, Greenspan H (2016) Longitudinal multiple sclerosis lesion segmentation using multiview convolutional neural networks. In: Proceedings of the deep learning in medical image analysis (DLMIA). Lecture notes in computer science, vol 10008, pp 58–67. https://doi.org/10. 1007/978-3-319-46976-8_7 Brosch T, Tang LY, Yoo Y, Li DK, Traboulsee A, Tam R (2016) Deep 3D convolutional encoder networks with shortcuts for multiscale feature integration applied to multiple sclerosis lesion segmentation. IEEE Trans Med Imaging 35(5):1229–1239. https://doi.org/10.1109/TMI.2016. 2528821 Buza K, Nagy GI, Nanopoulos A (2014) Storageoptimizing clustering algorithms for highdimensional tick data. Expert Syst Appl 41:4148–4157 Carreira J, Sminchisescu C (2012) CPMC: automatic object segmentation using constrained parametric min-cuts. In: PAMI Center Berkeley (2016) Caffe. http://caffe.berkeleyvision.org/ Chang H, Han J, Zhong C, Snijders A, Mao J-H (2017) Unsupervised transfer learning via multiscale convolutional sparse coding for biomedical applications. IEEE Trans Pattern Anal Mach Intell https://doi.org/10.1109/TPAMI.2017.2656884 Charbonnier J-P, van Rikxoort EM, Setio AAA, Schaefer-Prokop CM, van Ginneken B, Ciompi F (2017) Improving airway segmentation in computed tomography using leak detection with
116
A. Hazra et al.
convolutional networks. Med Image Anal 36:52–60. ISSN 1361-8415. https://doi.org/10.1016/j. media.2016.11.001 Chen X (2010) Google big table Chen W, Corso JJ (2015) Action detection by implicit intentional motion clustering. In: Proceedings of the IEEE international conference on computer vision, pp 3298–3306 Chen J, Chen Y, Xiaoyong D, Li C, Jiaheng L, Zhao S, Zhou X (2013) Big data challenge: a data management perspective. Front Comput Sci 7(2):157–164 Chen H, Dou Q, Yu L, Heng P-A (2016) Voxresnet: deep voxelwise residual networks for volumetric brain segmentation. arXiv:1608.05895 Chen S, Qin J, Ji X, Lei B, Wang T, Ni D, Cheng J-Z (2017) Automatic scoring of multiple semantic attributes with multi-task feature leverage: a study on pulmonary nodules in CT images. IEEE Trans Med Imaging 36(3):802–804. https://doi.org/10.1109/TMI.2016.2629462 Choi H, Jin KH (2016) Fast and robust segmentation of the striatum using deep convolutional neural networks. J Neurosci Methods 274:146–153. https://doi.org/10.1016/j.jneumeth.2016.10.007 Choudhary P, Hazra A (2019) Chest disease radiography in twofold: using convolutional neural networks and transfer learning. Evol Syst 1–13 Christodoulidis S, Anthimopoulos M, Ebner L, Christe A, Mougiakakou S (2017) Multisource transfer learning with convolutional neural networks for lung pattern analysis. IEEE J Biomed Health Inform 21(1):76–84. https://doi.org/10.1109/JBHI.2016.2636929 Çiçek Ö, Abdulkadir A, Lienkamp SS, Brox T, Ronneberger O (2016) 3D U-Net: learning dense volumetric segmentation from sparse annotation. In: Proceedings of the medical image computing and computer-assisted intervention. Lecture notes in computer science, vol 9901. Springer, pp 424–432. arXiv:1606.06650v1 Dalal N, Triggs B (2005) Histograms of oriented gradients for human detection. In: CVPR de Brebisson A, Montana G (2015) Deep neural networks for anatomical brain segmentation. In: Proceedings of the computer vision and pattern recognition, pp 20–28 Delakis M, Garcia C (2008) Text detection with convolutional neural networks. In: VISAPP Deng J, Dong W, Socher R, Li L, Li K, Li F (2009) Imagenet: a large-scale hierarchical image database. In: CVPR Deng L, Yu D et al (2014) Deep learning: methods and applications. Found Trends R Signal Process 7(3–4):197–387 Dittrich J, Blunschi L, Salles MAV (2011) Movies: indexing moving objects by shooting index images. Geoinformatica 15(4):727–767 dos Santos CN, Gatti M (2014) Deep convolutional neural networks for sentiment analysis for short texts. In: Proceedings of the international conference on computational linguistics (COLING 2014) Dou Q, Chen H, Yu L, Qin J, Heng PA (2017) Multilevel contextual 3-D CNN’s for false positive reduction in pulmonary nodule detection. IEEE Trans Biomed Eng 64(7):1558–1567. https://doi. org/10.1109/tbme.2016.2613502 Emad O, Yassine IA, Fahmy AS (2015) Automatic localization of the left ventricle in cardiac MRI images using deep learning. In: Proceedings of the IEEE engineering in medicine and biology society, pp 683–686. https://doi.org/10.1109/embc.2015.7318454 Everingham M, Eslami SA, Van Gool L, Williams CK, Winn J, Zisserman A (2014) The pascal visual object classes challenge: a retrospective. In: IJCV Felzenszwalb PF, Girshick RB, McAllester D, Ramanan D (2010) Object detection with discriminatively trained part-based models. In: PAMI Gao M et al (2016) Segmentation label propagation using deep convolutional neural networks and dense conditional random field. In: 2016 IEEE 13th international symposium on biomedical imaging (ISBI), Prague, pp 1265–1268. https://doi.org/10.1109/ISBI.2016.7493497 Gao X, Lin S, Wong TY (2015) Automatic feature learning to grade nuclear cataracts based on deep learning. IEEE Trans Biomed Eng 62(11):2693–2701. https://doi.org/10.1109/TBME.2015.244 4389
Recent Advances in Deep Learning Techniques …
117
Gers FA, Schmidhuber J, Cummins F (2000) Learning to forget: continual prediction with LSTM. In: Neural computation Ghafoorian M, Karssemeijer N, Heskes T, van Uden I, Sanchez C, Litjens G, de Leeuw F-E, van Ginneken B, Marchiori E, Platel B (2016a) Location sensitive deep convolutional neural networks for segmentation of white matter hyperintensities. arXiv:1610.04834 Ghafoorian M, Karssemeijer N, Heskes T, van Uden IWM, de Leeuw F-E, Marchiori E, van Ginneken B, Platel B (2016b) Non-uniform patch sampling with deep convolutional neural networks for white matter hyperintensity segmentation. In: Proceedings of the IEEE international symposium on biomedical imaging, pp 1414–1417. https://doi.org/10.1109/isbi.2016.749 3532 Gidaris S, Komodakis N (2015) Object detection via a multi-region and semantic segmentationaware CNN model. In: ICCV Glorot X, Bordes A, Bengio Y (2011) Domain adaption for large-scale sentiment classification: a deep learning approach. In: Proceedings of the international conference on machine learning (ICML 2011) Goodfellow IJ, Ibarz J, Arnoud S, Shet V (2014) Multi-digit number recognition from street view imagery using deep convolutional neural networks. In: ICLR Guan Z, Chen L, Zhao W, Zheng Y, Tan S, Cai D (2016) Weakly-supervised deep learning for customer review sentiment classification. In: Proceedings of the international joint conference on artificial intelligence (IJCAI 2016) Guggilla C, Miller T, Gurevych I (2016) CNN-and LSTM-based claim classification in online user comments. In: Proceedings of the international conference on computational linguistics (COLING 2016) Gulshan V, Peng L, Coram M, Stumpe MC, Wu D, Narayanaswamy A, Venugopalan S, Widner K, Madams T, Cuadros J, Kim R, Raman R, Nelson PC, Mega JL, Webster DR (2016) Development and validation of a deep learning algorithm for detection of diabetic retinopathy in retinal fundus photographs. J Am Med Assoc 316:2402–2410. https://doi.org/10.1001/jama.2016.17216 Han H, Wen Y, Chua T-S, Li X (2014) Toward scalable systems for big data analytics: a technology tutorial. IEEE Access 2:652–687 He K, Zhang X, Ren S, Sun J (2015a) Spatial pyramid pooling in deep convolutional networks for visual recognition. In: PAMI He P, Huang W, Qiao Y, Loy CC, Tang X (2015b) Reading scene text in deep convolutional sequences. In: CoRR He K, Zhang X, Ren S, Sun J (2015c) Deep residual learning for image recognition. In: CVPR Hinton GE, Salakhutdinov RR (2006) Reducing the dimensionality of data with neural networks. Science 313(5786):504–507 Hong S, You T, Kwak S, Han B (2015) Online tracking by learning discriminative saliency map with convolutional neural network. In: ICML Huang M, Qian Q, Zhu X (2017) Encoding syntactic knowledge in neural networks for sentiment classification. ACM Trans Inf Syst Hubel DH, Wiesel TN (1968) Receptive fields and functional architecture of monkey striate cortex. J Physiol Hwang S, Kim H-E, Jeong J, Kim H-J (2016) A novel approach for tuberculosis screening based on deep convolutional neural networks. In: Proceedings of the SPIE, medical imaging 2016: computer-aided diagnosis, vol 9785, p 97852W, 24 Mar 2016. https://doi.org/10.1117/12.221 6198 Jaderberg M, Vedaldi A, Zisserman A (2014) Deep features for text spotting. In: ECCV Jaderberg M, Simonyan K, Vedaldi A, Zisserman A (2015) Reading text in the wild with convolutional neural networks. In: IJCV Jafari M, Wang J, Qin Y, Gheisari M, Shahabi AS, Tao X (2016) Automatic text summarization using fuzzy inference. In: 2016 22nd international conference on automation and computing (ICAC). IEEE, pp 256–260
118
A. Hazra et al.
Janowczyk A, Basavanhally A, Madabhushi A (2017) Stain normalization using sparse autoencoders (STANOSA): application to digital pathology. Comput Med Imaging Graph 57:50–61 Johnson R, Zhang T (2015) Effective use of word order for text categorization with convolutional neural networks. In: Proceedings of the conference of the North American chapter of the association for computational linguistics: human language technologies (NAACL-HLT 2015) Kalchbrenner N, Grefenstette E, Blunsom P (2014) A convolutional neural network for modelling sentences. In: Proceedings of the annual meeting of the association for computational linguistics (ACL 2014) Kim Y (2014) Convolutional neural networks for sentence classification. In: Proceedings of the annual meeting of the association for computational linguistics (ACL 2014) Kim H, Hwang S (2016) Scale-invariant feature Learning using deconvolutional neural networks for weakly-supervised semantic segmentation. arXiv:1602.04984 Krause J, Jin H, Yang J, Fei-Fei L (2015) Fine-grained recognition without part annotations. In: CVPR Krizhevsky A, Sutskever I, Hinton GE (2012) Imagenet classification with Deep Convolutional neural networks. In: NIPS Lawrence S, Giles CL, Tsoi AC, Back AD (1997) Face recognition: a convolutional neural-network approach. In: TNN Le Cun BB, Denker JS, Henderson D, Howard RE, Hubbard W, Jackel LD (1990) Handwritten digit recognition with a back-propagation network. In: NIPS Le Q, Mikolov T (2014) Distributed representations of sentences and documents. In: Proceedings of the international conference on machine learning (ICML 2014) Li G, Ooi BC, Feng J, Wang J, Zhou L (2008) Ease: an effective 3-in-1 keyword search method for unstructured, semi-structured and structured data. In: Proceedings of the 2008 ACM SIGMOD international conference on management of data. ACM, pp 903–914 Li H, Li Y, Porikli F (2014) Deeptrack: learning discriminative feature representations by convolutional neural networks for visual tracking. In: BMVC Li W, Jia F, Hu Q (2015) Automatic segmentation of liver tumor in CT images with deep convolutional neural networks. J Comput Commun 3(11):146–151 Li Z, Zhang Y, Wei Y, Wu Y, Yang Q (2017) End-to-end adversarial memory network for crossdomain sentiment classification. In: Proceedings of the international joint conference on artificial intelligence (IJCAI 2017) Lin T-Y, Maire M, Belongie S, Hays J, Perona P, Ramanan D, Dollár P, Zitnick CL (2014) Microsoft COCO: common objects in context. In: ECCV (2014) Lin T-Y, RoyChowdhury A, Maji S (2015a) Bilinear cnn models for fine-grained visual recognition. arXiv:1504.07889 Lin D, Shen X, Lu C, Jia J (2015b) Deep LAC: Deep localization, alignment and classification for fine-grained recognition. In: CVPR Liu J, Zhang Y (2017) Attention modeling for targeted sentiment. In: Proceedings of the conference of the European chapter of the association for computational linguistics (EACL 2017) Liu W, Anguelov D, Erhan D, Szegedy C, Reed S (2015) SSD: single shot multibox detector. In: ECCV Lo SCB, Lou SLA, Lin J-S, Freedman MT, Chien MV, Mun SK (1995) Artificial convolution neural network techniques and applications for lung nodule detection. IEEE Trans Med Imaging 14(4):711–718. https://doi.org/10.1109/42.476112 Loshchilov I, Hutter F (2016) Online batch selection for faster training of neural networks. In: ICLR workshop Lu Y, Javidi T, Lazebnik S (2016a) Adaptive object detection using adjacency and zoom prediction. In: CVPR Lu X, Xu D, Liu D (2016b) Robust 3D organ localization with dual learning architectures and fusion. In: Proceedings of the deep learning in medical image analysis (DLMIA). Lecture notes in computer science, vol 10008, pp 12–20. https://doi.org/10.1007/978-3-319-46976-8_2
Recent Advances in Deep Learning Techniques …
119
Masci J, Meier U, Cire¸san D, Schmidhuber J (2011) Stacked convolutional auto-encoders for hierarchical feature extraction. In: Proceedings of the international conference on artificial neural networks, pp 52–59 McAfee A, Brynjolfsson E, Davenport TH, Patil DJ, Barton D (2012) Big data. The management revolution. Harv Bus Rev 90(10):61–67 Michael K, Miller KW (2013) Big data: new opportunities and new challenges [guest editors’ introduction]. Computer 46(6):22–24 Microsoft (2016) CNTK. https://github.com/Microsoft/CNTK Mikolov T, Chen K, Corrado G, Dean J (2013a) Efficient estimation of word representations in vector space. In: Proceedings of international conference on learning representations (ICLR 2013) Mikolov T, Sutskever I, Chen K, Corrado G, Dean J (2013b) Distributed representations of words and phrases and their compositionality. In: Proceedings of the annual conference on advances in neural information processing systems (NIPS 2013) Mishra A, Dey K, Bhattacharyya P (2017) Learning cognitive features from gaze data for sentiment and sarcasm classification using convolutional neural network. In: Proceedings of the annual meeting of the association for computational linguistics (ACL 2017) Mnih A, Kavukcuoglu K (2013) Learning word embeddings efficiently with noise-contrastive estimation. In: Proceedings of the annual conference on advances in neural information processing systems (NIPS 2013) Moraes R, Valiati JF, Neto WP (2013) Document-level sentiment classification: an empirical comparison between SVM and ANN. Expert Syst Appl Morin F, Bengio Y (2005) Hierarchical probabilistic neural network language model. In: Proceedings of the international workshop on artificial intelligence and statistics Ngo TA, Lu Z, Carneiro G (2017) Combining deep learning and level set for the automated segmentation of the left ventricle of the heart from cardiac cine magnetic resonance. Med Image Anal 35:159–171. https://doi.org/10.1016/j.media.2016.05.009 Nilsback M-E, Zisserman A (2008) Automated flower classification over a large number of classes. In: ICVGIP Philip Chen CL, Zhang C-Y (2014) Data-intensive applications, challenges, techniques and technologies: a survey on big data. Inf Sci 275:314–347 Plis SM, Hjelm DR, Salakhutdinov R, Allen EA, Bockholt HJ, Long JD, Johnson HJ, Paulsen JS, Turner JA, Calhoun VD (2014) Deep learning for neuroimaging: a validation study. Front Neurosci. https://doi.org/10.3389/fnins.2014.00229 Pluim JP, Maintz JA, Viergever M et al (2003) Mutual-information-based registration of medical images: a survey. In: T-MI Porkar P (2012) Sensor networks challenges. In: 11th international conference on data networks, DNCOCO ’12, 7–9 Sept 2012 Poudel RPK, Lamata P, Montana G (2016) Recurrent fully convolutional neural networks for multislice MRI cardiac segmentation. arXiv:1608.03974 Poultney C et al (2006) Efficient learning of sparse representations with an energy-based model. In: Proceedings of the advances in neural information processing systems, pp 1137–1144 Prasoon A, Petersen K, Igel C, Lauze F, Dam E, Nielsen M (2013) Deep feature learning for knee cartilage segmentation using a triplanar convolutional neural network. In: Proceedings of the medical image computing and computer-assisted intervention. Lecture notes in computer science, vol 8150, pp 246–253. https://doi.org/10.1007/978-3-642-40763-5_31 Qian Q, Tian B, Huang M, Liu Y, Zhu X, Zhu X (2015) Learning tag embeddings and tag-specific composition functions in the recursive neural network. In: Proceedings of the annual meeting of the association for computational linguistics (ACL 2015) Qian Q, Huang M, Lei J, Zhu X (2017) Linguistically regularized LSTM for sentiment classification. In: Proceedings of the annual meeting of the association for computational linguistics (ACL 2017) Rajkumar A, Lingam S, Taylor AG, Blum M, Mongan J (2017) High-throughput classification of radiographs using deep convolutional neural networks. J Digit Imaging 30:95–101. https://doi. org/10.1007/s10278-016-9914-9
120
A. Hazra et al.
Redmon J, Divvala S, Girshick R, Farhadi A (2016) You only look once: unified, real-time object detection. In: CVPR Riabov A, Liu Z (2006) Scalable planning for distributed stream processing systems. In: ICAPS, pp 31–41 Rifai S, Vincent P, Muller X, Glorot X, Bengio Y (2011) Contractive auto-encoders: explicit invariance during feature extraction. In: Proceedings of the international conference on machine learning, pp 833–840 Rowley HA, Baluja S, Kanade T (1998) Neural network-based face detection. In: PAMI Salakhutdinov R, Hinton GE (2009) Deep Boltzmann machines. In: Proceedings of the international conference on artificial intelligence and statistics, vol. 1, art no 3 Sarraf S, Tofighi G (2016) Classification of Alzheimer’s disease using fMRI data and deep learning convolutional neural networks. arXiv:1603.08631 Sermanet P, Eigen D, Zhang X, Mathieu M, Fergus R, LeCun Y (2013) Overfeat: integrated recognition, localization and detection using convolutional networks Setio AAA et al (2016) Pulmonary nodule detection in CT images: false positive reduction using multi-view convolutional networks. IEEE Trans Med Imaging 35(5):1160–1169. https://doi.org/ 10.1109/TMI.2016.2536809 Shen W, Zhou M, Yang F, Yang C, Tian J (2015a) Multi-scale convolutional neural networks for lung nodule classification. In: Ourselin S, Alexander D, Westin CF, Cardoso M (eds) Information processing in medical imaging. IPMI 2015. Lecture notes in computer science, vol 9123. Springer, Cham Shen W, Yang F, Mu W, Yang C, Yang X, Tian J (2015b) Automatic localization of vertebrae based on convolutional neural networks. In: Proceedings of the SPIE on medical imaging, vol 9413, p 94132E Shi B, Bai X, Yao C (2015) An end-to-end trainable neural network for image-based sequence recognition and its application to scene text recognition. In: CoRR Siddiqa A, Hashem IAT, Yaqoob I, Marjani M, Shamshirband S, Gani A, Nasaruddin F (2016) A survey of big data management: taxonomy and state-of-the-art. J Netw Comput Appl 71:151–166 Simard PY, Steinkraus D, Platt JC (2003) Best practices for convolutional neural networks applied to visual document analysis. In: ICDAR Simonyan K, Zisserman A (2015) Very deep convolutional networks for large-scale image recognition. In: ICLR Stollenga MF, Byeon W, Liwicki M, Schmidhuber J (2015) Parallel multi-dimensional LSTM, with application to fast biomedical volumetric image segmentation. In: Proceedings of the advances in neural information processing systems, pp 2998–3006 Suk H-I, Shen D (2016) Deep ensemble sparse regression network for Alzheimer’s disease diagnosis. In: Proceedings of the medical image computing and computer-assisted intervention. Lecture notes in computer science, vol 10019, pp 113–121. https://doi.org/10.1007/978-3-319-471570_14 Sun W, Zheng B, Qian W (2016) Computer-aided lung cancer diagnosis with deep learning algorithms. In: Proceedings of the SPIE, medical imaging 2016: computer-aided diagnosis, vol 9785, p 97850Z, 24 Mar 2016. https://doi.org/10.1117/12.2216307 Suzani A, Rasoulian A, Seitel A, Fels S, Rohling R, Abolmaesumi P (2015) Deep learning for automatic localization, identification, and segmentation of vertebral bodies in volumetric MR images. In: Proceedings of the SPIE medical imaging, vol 9415, p. 941514 Tang D, Qin B, Liu T (2015) Document modelling with gated recurrent neural network for sentiment classification. In: Proceedings of the conference on empirical methods in natural language processing (EMNLP 2015) Tang D, Qin B, Feng X, Liu T (2016) Effective LSTMs for target-dependent sentiment classification. In: Proceedings of the international conference on computational linguistics (COLING 2016) Tarando SR, Fetita C, Faccinetto A, Brillet P-Y (2016) Increasing CAD system efficacy for lung texture analysis using a convolutional network. In: Proceedings of the SPIE, medical imaging
Recent Advances in Deep Learning Techniques …
121
2016: computer-aided diagnosis, vol 9785, p 97850Q, 24 Mar 2016. https://doi.org/10.1117/12. 2217752 Tran PV (2016) A fully convolutional neural network for cardiac segmentation in short-axis MRI. arXiv:1604.00494 Uijlings JR, van de Sande KE, Gevers T, Smeulders AW (2013) Selective search for object recognition. In: IJCV Vaillant R, Monrocq C, Le Cun Y (1994) Original approach for the localisation of objects in images. In: IEE proceedings—vision, image and signal processing van Grinsven MJJP, van Ginneken B, Hoyng CB, Theelen T, Sánchez CI (2016) Fast convolutional neural network training using selective data sampling: application to hemorrhage detection in color fundus images. IEEE Trans Med Imaging 35(5):1273–1284. https://doi.org/10.1109/TMI. 2016.2526689 van Tulder G, de Bruijne M (2016) Combining generative and discriminative representation learning for lung CT analysis with convolutional restricted Boltzmann machines. IEEE Trans Med Imaging 35(5):1262–1272. https://doi.org/10.1109/TMI.2016.2526687 Vincent P, Larochelle H, Bengio Y, Manzagol P-A (2008) Extracting and composing robust features with denoising autoencoders. In: Proceedings of the international conference on machine learning, pp 1096–1103 Vivanti R, Ephrat A, Joskowicz L, Karaaslan O, Lev-Cohain N, Sosna J (2015) Automatic liver tumor segmentation in follow-up CT studies using convolutional neural networks. In: Proceedings of the patch-based methods in medical image processing workshop, MICCAI’2015, pp 54–61 Waibel A, Hanazawa T, Hinton G, Shikano K, Lang KJ (1990) Phoneme recognition using timedelay neural networks. In: Readings in speech recognition, pp 393–404 Wang X, Gupta A (2015) Unsupervised learning of visual representations using videos. In: ICCV Wang Z, Wang X, Wang G (2015a) Learning fine-grained features via a cnn tree for large-scale classification. arXiv:1511.04534 Wang X, Liu Y, Sun C, Wang B, Wang X (2015c) Predicting polarities of tweets by composing word embeddings with long short-term memory. In: Proceedings of the annual meeting of the association for computational linguistics (ACL 2015) Wang J, Yu L-C, Lai RK, Zhang X (2016b) Dimensional sentiment analysis using a regional CNN-LSTM model. In: Proceedings of the annual meeting of the association for computational linguistics (ACL 2016) Wang X, Jiang W, Luo Z (2016c) Combination of convolutional and recurrent neural network for sentiment analysis of short texts. In: Proceedings of the international conference on computational linguistics (COLING 2016) Wang S, Yao J, Xu Z, Huang J (2016e) Subtype cell detection with an accelerated deep convolution neural network. In: Proceedings of the medical image computing and computer-assisted intervention. Lecture notes in computer science, vol 9901, pp 640–648. https://doi.org/10.1007/9783-319-46723-8_74 Wang C, Elazab A, Wu J, Hu Q (2017) Lung nodule classification using deep feature fusion in chest radiography. Comput Med Imaging and Graph 57:10–18. ISSN 0895-6111. https://doi.org/10. 1016/j.compmedimag.2016.11.004 Williams RJ, Zipser D (1989) A learning algorithm for continually running fully recurrent neural networks. Neural Comput 1(2):270–280 Witten IH, Frank E, Hall MA, Pal CJ (2016) Data mining: practical machine learning tools and techniques. Morgan Kaufmann Xiao T, Zhang J, Yang K, Peng Y, Zhang Z (2014) Error-driven incremental learning in deep convolutional neural network for large-scale image classification. In: ACMMM Xie W, Noble JA, Zisserman A (2016) Microscopy cell counting and detection with fully convolutional regression networks. Comput Methods Biomech Biomed Eng Imaging Vis 1–10 Xu H, Su F (2015) Robust seed localization and growing with deep convolutional features for scene text detection. In: ICMR
122
A. Hazra et al.
Xu J, Luo X, Wang G, Gilmore H, Madabhushi A (2016a) A deep convolutional neural network for segmenting and classifying epithelial and stromal regions in histopathological images. Neurocomputing 191:214–223. https://doi.org/10.1016/j.neucom.2016.01.034 Xu Y, Li Y, Liu M, Wang Y, Lai M, Chang EI-C (2016b) Gland instance segmentation by deep multichannel side supervision. arXiv:1607.03222 Yan Z, Jagadeesh V, DeCoste D, Di W, Piramuthu R (2014) HD-CNN: hierarchical deep convolutional neural network for image classification. arXiv:1410.0736 Yang L, Luo P, Loy CC, Tang X (2015) A large-scale car dataset for fine-grained categorization and verification. In: CVPR Yang W, Chen Y, Liu Y, Zhong L, Qin G, Lu Z, Feng Q, Chen W (2017a) Cascade of multi-scale convolutional neural networks for bone suppression of chest radiographs in gradient domain. Med Image Anal 35:421–433. ISSN 1361-8415. https://doi.org/10.1016/j.media.2016.08.004 Yang M, Tu W, Wang J, Xu F, Chen X (2017b) Attention-based LSTM for target-dependent sentiment classification. In: Proceedings of AAAI conference on artificial intelligence (AAAI 2017) Yoo D, Park S, Lee J-Y, Paek AS, So Kweon I (2015) Attentionnet: Aggregating weak directions for accurate object detection. In: CVPR Younes L (1999) On the convergence of markovian stochastic algorithms with rapidly decreasing ergodicity rates. Stoch: Int J. Probab Stoch Process 65(3/4):177–228 Yu A, Grauman K (2014) Fine-grained visual comparisons with local learning. In: CVPR Yu J, Jiang J (2016) Learning sentence embeddings with auxiliary tasks for cross-domain sentiment classification. In: Proceedings of the conference on empirical methods in natural language processing (EMNLP 2016) Yu L, Yang X, Chen H, Qin J, Heng PA (2017) Volumetric convnets with mixed residual connections for automated prostate segmentation from 3D MR images. In: Proceedings of the thirty-first AAAI conference on artificial intelligence Zhang W, Li R, Deng H, Wang L, Lin W, Ji S, Shen D (2015a) Deep convolutional neural networks for multi-modality isointense infant brain image segmentation. Neuroimage 108:214–224. https:// doi.org/10.1016/j.neuroimage.2014.12.061 Zhang C, Yao C, Shi B, Bai X (2015b) Automatic discrimination of text and non-text natural images. In: ICDAR Zhao J, Zhang M, Zhou Z, Chu J, Cao F (2016) Automatic detection and classification of leukocytes using convolutional neural networks. Med Biol Eng Comput. https://doi.org/10.1007/s11517016-1590-x Zhu Y, Wang L, Liu M, Qian C, Yousuf A, Oto A, Shen D (2017) MRI Based prostate cancer detection with high-level representation and hierarchical classification. Med Phys 44(3):1028–1039. https:// doi.org/10.1002/mp.12116 Zilly J, Buhmann JM, Mahapatra D (2017) Glaucoma detection using entropy sampling and ensemble learning for automatic optic cup and disc segmentation. Comput Med Imaging Graph 55:28–41. https://doi.org/10.1016/j.compmedimag.2016.07.012
Validation of a New Method of Pediatric Refraction: Large Aperture Lens Rack Anupam Sahu, Samrat Chatterjee, Deepshikha Agrawal, and Pradeep Chand Dubey
Abstract Purpose: To minimize the problems faced during pediatric retinoscopy, a modification to the standard lens rack was devised at our institute. This study aims to validate the refractive errors determined by large aperture lens rack (LALR) as compared to the results with trial lens method and analyze time taken for trial lenses versus LALR. Methods: It was a single center, prospective, pilot study. New patients in the age group of 5 to 13 years with decreased vision, improving to 20/20 with pinhole were included. Cycloplegicretinoscopy was done with both trial lenses and LALR for all patients. Post mydriatictest(PMT) was done at 1 week. Paired T-test was used to compare groups. Correlations were observed using Pearson’s 2 tailed correlation. Results: Mean refractive error in spherical equivalent was –0.95D with trial lens and –0.65D with LALR (p = 0.07). Mean PMT value was –1.0D. Mean cylinder measured with trial lens was –1.48D and –1.26D with LALR (p = 0.06). Mean cylinder value at PMT was 1.16D. Trial lens value was significantly different from PMT value (p < 0.01). Time taken for LALR was 12 s more than trial lenses (p = 0.58). Both spherical equivalent and cylindrical values were significantly correlating with each other for both groups (p < 0.01). Conclusions: The results obtained by LALR were similar to retinoscopy done with trial lenses for the total refraction. LALR was more accurate in estimating the cylindrical values. We expect time taken for LALR to decrease further with repeated usage. Keywords Retinoscopy · Lens rack · Refraction · Pediatric
1 Introduction Refraction among preverbal children has always been difficult. Various attempts have been made in an attempt to quicken the procedure while retaining reliability, including the use of automated refractometers Wood (1987). However manual retinoscopy under cycloplegia remains the gold standard of pediatric refraction Guha (2017), A. Sahu (B) · S. Chatterjee · D. Agrawal · P. C. Dubey MGM Eye Institute, Raipur, Chhattisgarh, India e-mail: [email protected] © Springer Nature Singapore Pte Ltd. 2021 A. A. Rizvanov et al. (eds.), Advances in Biomedical Engineering and Technology, Lecture Notes in Bioengineering, https://doi.org/10.1007/978-981-15-6329-4_11
123
124
A. Sahu et al.
Erdurmus et al. (2007), Ozdemir et al. (2015). Cycloplegicretinoscopy requires time, patience, and expertise of the examiner along with cooperation of the child. A retinoscopy rack (ret rack) is an equipment that contains many different spherical lenses in different powers. However the commonly available ret rack has a reduced aperture size which makes it difficult to use in children due to continuous eye movements. To minimize the problems faced during pediatric retinoscopy, a modification to the standard lens rack was devised at our institute. The idea was to reduce the redundancy of placing trial lenses by developing a product where all/nearly all the lenses (plus or minus) of various powers will be in one lens rack. Also, it should be quick and easy to carry. The lens rack devised consists of rectangular-shaped hard-coated CR39 lenses (30 × 18 mm) which are easily available in the market, mounted on to a wooden rack (Fig. 1). These are light weight, non-allergic, and easy to handle. There are 4 lens racks for plus powers and 4 lens racks for minus powers which offer a range of ±20 D. Each lens rack has six lenses in ascending order of power. This study aims to validate the refractive errors determined by large-aperture lens rack (LALR) as compared to the results with trial lens method and analyze time taken for trial lenses versus LALR.
Fig. 1 The large aperture lens rack
Validation of a New Method of Pediatric …
125
2 Methods It was a single center, prospective, pilot study from 16 February 2017 to 24 March 2017 at MGM Eye Institute, Raipur. Institutional Ethics Committee approval was obtained and the study was conducted in full accord with the tenets of the Declaration ofHelsinki. New patients in the age group of 5 to 13 years with decreased vision (less than 20/20 unaided), improving to 20/20 on LogMAR chart with pinhole were included. Children or eyes with amblyopia or other pathological causes in which vision did not improve to 20/20 were excluded. Also, children who did not cooperate for the post mydriatic test at 1 week were excluded. A written informed consent was taken from all parents/guardians before enrolling. Consent was also obtained for the use of photographs contained in the medical record for the purpose of publication. Once enrolled it was ensured that none of the senior optometrists selected for the refraction did the comprehensive eye examination pre-dilatation. Cycloplegicretinoscopy was done with both trial lenses and LALR by separate senior optometrists who were randomized by lottery (Figs. 2 and 3). Post mydriatic test (PMT) is an assessment of the findings of cycloplegic refraction by subjective means after the effect of cycloplegia is eliminated. It was done at 1 week again by lottery randomization. The values of refractive error (in diopters) in terms of spherical equivalent (spherical amount + ½ cylindrical amount) and cylindrical value were collected and entered in SPSS (Version 16) for LALR, trial lenses and PMT, and compared using a paired T-test. Time taken for retinoscopy was recorded with a stop watch by a neutral person for both trial lenses and LALR and the average time taken was then compared. These
Fig. 2 Retinoscopy performed with LALR
126
A. Sahu et al.
Fig. 3 Retinoscopy performed with trial lenses
were compared using paired T-test with p value < 0.05 considered significant. Correlations were observed using Pearson’s 2 tailed correlation and p value < 0.01 was considered significant.
3 Results 28 eyes of 15 patients whose parents consented for the trial and who completed all steps were examined. 2 eyes with amblyopia were excluded. 53.33% (n = 8) were male. The mean age was 8.8 ± 2.5 years.
3.1 Comparison Between Retinoscopy Values Mean refractive error in spherical equivalent (SE) was –0.95 ± 2.3 Diopter (D) with trial lens and –0.65 ± 2.5D with LALR. Though the mean dioptric difference was 0.3D, paired T-test showed that this difference was not statistically significant (p = 0.07). Mean PMT value in SE was –1.0 ± 1.8D. This was closer to the value achieved with trial lens. But this difference was also statistically insignificant with trial lens having a p value of 0.76 when paired with PMT value and LALR having a p value of 0.11. Mean cylinder measured with trial lens was –1.48 ± 1.4D, whereas it was –1.26 ± 1.3D with LALR. Both had no statistically significant difference (p = 0.06). The mean cylinder value at PMT was 1.16 ± 1.2D. This was closer to the LALR values (p
Validation of a New Method of Pediatric … Table 1 Comparison of retinoscopy values between groups
127
Factors assessed Trial lens group LALR group PMT value Mean refractive −0.95 D error in SE
−0.65 D
−1.0 D
Mean cylinder measured
−1.48 D
−1.26 D
−1.16 D
Time taken
152 s
164 s
NA
= 0.21) with trial lens retinoscopy overestimating the cylindrical value significantly (p < 0.01). Time taken in retinoscopy by trial lens was 152 ± 63.6 s for each patient whereas time taken by LALR was 164 ± 57.7 s. Time taken for retinoscopy was evaluated in 15 patients. Though LALR took 12 s more for the same set of patients, the difference was not significant (paired T-test; p = 0.58) (Table 1).
3.2 Correlation Between Retinoscopy Values To evaluate whether the values were correlating through the entire spectrum of refractive errors studied, Pearson’s correlation was used. Both spherical equivalent and cylindrical values showed strong linear correlation with each other for both groups and their correlation was similar with PMT values (p < 0.01).
4 Discussion Various attempts have been made to get reproducible refractive error calculations in children. A significant focus is on automated refractometers as they are quick and can be done by non-clinical persons. However, even after multiple instruments, it has found acceptability only as a screening device for the field where non-clinical persons can carry out the retinoscopy within a short span of time over a large number of childrenGuha (2017); Erdurmus et al. 2007; Ozdemir et al. 2015). However, cycloplegic refraction is still required in those preverbal children who fail the screening testMorgan et al. 2015). In our short trial, the large aperture lens rack provided comparable results to trial lens retinoscopy. In terms of astigmatism, it was significantly closer to the final PMT values, indicating that the large aperture may provide a better window to evaluate the retinoscope reflex. While it was expected to shorten the time of retinoscopy, LALR in fact took slightly longer. We presume it could be because of a learning curve for the new instrument, though our optometrists were using it for a few weeks before the study. We aim to try it out with our newer trainees for feedback.
128
A. Sahu et al.
4.1 Limitations Our study has several limitations. First, this was a pilot study to assess the comfort of our optometrists with the device and these results should not be taken as clinically relevant at this point of the study. Second, we may not have been able to nullify the inter and intraobserver differences as noticed for cycloplegicretinoscopy (McCullough et al. 2017).
5 Conclusion The results obtained by LALR were similar to retinoscopy done with trial lenses for the total refraction. Cylinder values for astigmatism measured by LALR were significantly closer to the PMT values. We expect time taken for the procedure to decrease further with repeated usage. A larger study, taking into account inter and an intra observer difference has been planned. Overall, the LALR provides a lightweight device which can be carried even to more remote locations, unlike the whole retinoscopy trial set. It also provides a large aperture for better appreciation of astigmatism. It is a much cheaper alternative to automated refractors and can be made with locally available materials.
References Erdurmus M, Yagci R, Karadag R, Durmus M (2007) A comparison of photorefraction and retinoscopy in children. J Am Assoc Pediatr Ophthalmol Strabismus 11:606–611 Guha S et al (2017) A comparison of cycloplegic autorefraction and retinoscopy in Indian children. Clin Exp Optom 100:73–78 Morgan IG, Iribarren R, Fotouhi A, Grzybowski A (2015) Cycloplegic refraction is the gold standard for epidemiological studies. Acta Ophthalmol (Copenh) 93:581–585 McCullough SJ, Doyle L, Saunders KJ (2017) Intra- and inter- examiner repeatability of cycloplegic retinoscopy among young children. Ophthalmic Physiol Opt J Br Coll Ophthalmic Opt Optom 37:16–23 Ozdemir O, Özen Tunay Z, Petriçli IS, Ergintürk Acar D, Erol MK (2015) Comparison of noncycloplegic photorefraction, cycloplegic photorefraction and cycloplegic retinoscopy in children. Int J Ophthalmol 8:128–131 Wood ICJ (1987) A review of autorefractors. Eye 1:529–535
Comparative Evaluation of in Vitro Antioxidant, Amylase Inhibition and Cytotoxic Activity of Cur-Pip Dual Drug Loaded Nanoparticles Trilochan Satapathy, Prasanna Kumar Panda, and Gitanjali Mishra
Abstract Objective: To comparatively evaluate in vitro antioxidant, amylase inhibition, and cytotoxic potential of Cur-Pip dual drug incorporated nanoparticles developed by nano co-precipitation method. Methods: Cur-nanoparticles, Pipnanoparticles, and Cur-Pip dual drug-loaded nanoparticles were developed by nano co-precipitation method. The amylase inhibition assays of formulated nanoparticles were performed by using the chromogenic DNSA method. The total antioxidant activity was performed by using Butylated hydroxyanisole (BHA) as reference standard. The cytotoxicity potential of nanoparticles formulations at different concentrations such as 25, 50, 100, 250, 500, and 1000 μg were evaluated by agar diffusion method using bacteria E. coli AB 1157. Dimethylsulfoxide (DMSO) was used as solvent. Results: The amylase inhibition assay of three different formulations were performed at concentrations 1000, 500, and 100 μg/ml. Out of these three formulations curcumin nanoparticles exhibited maximum amylase inhibition i.e. –404.92%. Total antioxidant capacity of different nanoformulations were evaluated at concentrations 10, 50, and 100 μg. Antioxidant potentials are expressed as equivalents of ascorbic acid. The Cur + Pip nanoparticles showed maximum antioxidant activity, i.e., 34.6. In cytotoxicity study, none of the formulations exhibited any zone of inhibition. Conclusions: The Cur + Pip nanoparticles showed maximum antioxidant activity in comparison with cur-nanoparticles as well as Pip-nanoparticles whereas cur-nanoparticles exhibited better amylase inhibition over other two nanoformulations. None of the developed formulation showed any cytotoxic activity. Keywords Nanoparticles · Antioxidant · Amylase inhibition · Cytotoxic
T. Satapathy (B) · G. Mishra Berhampur University, Bhanjabihar, Berhampur, Ganjam Odisha-760007, India e-mail: [email protected] P. K. Panda University Department of Pharmaceutical Sciences, Utkal University, Vanivihar, Bhubaneswar Odisha-751004, India © Springer Nature Singapore Pte Ltd. 2021 A. A. Rizvanov et al. (eds.), Advances in Biomedical Engineering and Technology, Lecture Notes in Bioengineering, https://doi.org/10.1007/978-981-15-6329-4_12
129
130
T. Satapathy et al.
1 Introduction Curcumin is a yellow polyphenol isolated from the rhizome of herb turmeric (Curcuma longa). Curcumin is not only used in India but also used in several countries like China and Southeast Asia as a spice. Exhaustive study has been carried out on curcumin by various researchers indicated that, Curcumin possess distinct pharmacological activities (Anand et al. 2008) and used in the treatment of various ailments such as anorexia, biliary disorder, hepatic disorder, wounds. (Aggarwal and Harikumar 2009) Further research demonstrated that, Curcumin alone or in combination with other antineoplastic agents exhibited potent anticancer potential which was experimentally evidenced by its inhibitory effects observed on the growth and multiplication of in vitro cell lines as well as In-vivo melanoma, pancreatic carcinomas, Ovarian carcinoma, etc. (Aggarwal et al. 2003) Previous research by various researchers indicated that, Curcumin also possess exciting properties as a signaling molecule which down-regulates gene expression associated with angiogenesis, setoff apoptotic mechanism, and cell cycle arrest. (Gururaj et al. 2002; Belakavadi and Salimath 2005; Shishodia et al. 2005) Recently Jia et al. (2018) investigated the effects of curcumin nanoparticles in inhibition of diabetic neuropathic pain (DNP) and found that, the nanoparticles containing curcumin able to decrease the up-regulated P2Y12 expression in the dorsal root ganglia (DRG), reduced the up-regulation of IL-1β and Cx43, and decreased phosphorylated-Akt (p-Akt) in the dorsal root ganglia of rats with diabetes mellitus. Abdel-Mageid et al. (2018) also investigated the effectiveness of curcumin nanoparticles against complications associated with experimentally induced diabetes in rats, curcumin nanoparticles has greater therapeutic effectiveness in the treatment of Diabetic cardiomyopathy, This is due to by decrease the cardiac inflammation, myocardial fibrosis, as well as programmed myocardial cell deaths. Now-a-days Curcuminoids also gaining special attention in various cosmeceutical products. (Joshi and Pawar 2015) Despite having all these pharmacological properties, curcumin is not approved by the regulatory authorities as a therapeutic agent because of its stability issues (SundarDhilip Kumar et al. 2018). To overcome the problems several strategies have been encouraged such as encapsulation of curcumin in nanoparticles, microspheres, Liposomes, etc., to enhance the curcumin delivery. Extensive review of literature has been done and found that, the bioavailability of curcumin can be enhanced by the addition of piperine (bioenhancer) and by formulate as a nanoparticle. Development of formulations containing controlled drug delivery of therapeutic agents with improved stability and bioavailability is the main aspects of today’s research on which many research can be done. Polymeric nanoparticles are considered as a promising drug delivery system that can helpful to overcome the problems associated with Curcumin delivery with stability and enhanced bioavailability. Among the polymers used to formulate the nanoparticles, Polycaprolactone (PCL), Poly (lactic-co-glycolic acid) or PLGA, Poly(n-butyl cyanoacrylate) or PBCA, etc., are considered as ideal polymers because of their low toxicity, biodegradability. Hence, in the present research, our efforts have been devoted to formulate three different nanoformulations by using nanoprecipitation
Comparative Evaluation of in Vitro Antioxidant …
131
method and to screen these formulations for total antioxidant, amylase inhibition as well as bacterial-based cytotoxicity study.
2 Materials and Methods 2.1 Materials DNS solution: Onegm of DNS was dissolved in 2 N NaOH, and to it thirty gm of potassium sodium tartarate was added and the whole volume was made up to 100 ml. Amylase (Diastase procured from HiMedia, Mumbai, Cat No. RM-638), Butylated hydroxyl anisole (BHA), and other chemicals used in this study were pure analytical grade and obtained from commercial sources. These chemicals were used in experiment without any further purification. Curcumin and piperine were kind gift sample from Sunpure, New Delhi, India.
2.2 Methods Three different nanoparticle formulations were developed by nano co-precipitation method. The amylase inhibition potential of these nanoformulations was evaluated by chromogenic DNSA (3, 5-Dinitrosalicylic acid) method. The total antioxidant activity was performed by using Butylated hydroxyl anisole (BHA) as reference standard and cytotoxicity potential of different nanoparticle formulations at different concentrations were evaluated by agar diffusion method using bacteria E. coli AB 1157.
2.3 Preparation of Nanoparticles In this research, nanoparticles were developed by nanoprecipitation method. Three different formulations such as Cur-NP, Pip-NP, and Cur-Pip dual drug-loaded nanoparticles were developed by using polymer Chitosan/Polycaprolactone (PCL). Briefly, the pure drug (Curcumin/Piperine), polymer (Chitosan), and copolymer (PCL) were taken at their appropriate ratios and dissolved into 5 ml of (90%) acetic acid solution. The homogenous solution was formed due to continuous stirring which then added drop wise into a beaker containing distilled water (50 ml). Finally the nanoparticles formed which was subjected to centrifuge at 13,000 rpm/min for 30 min. After completion of the centrifugation process, the supernatant solution was discarded and the remaining content re-suspended with 10 ml of fresh distilled water. The nanoparticles were then subjected for further analysis.
132
T. Satapathy et al.
2.4 Amylase Inhibition Assay The amylase inhibition assay was performed by using the chromogenic DNSA method. (Miller 1959; Sudha et al. 2011) The total assay mixture consists of one thousand four hundred (1400) μl of 0.05 M sodium phosphate buffer (pH 6.9), Fifty (50) μl of amylase (Diastase procured from HiMedia, Mumbai, Cat No. RM-638) and samples at concentration thousand (1000), Five hundred (500) and hundred (100) μg/ml were incubated at 37 °C for ten (10) min. After pre-incubation, Five hundred (500) μl of 1% (w/v) starch solution was added in the above buffer to each tube and allowed to incubate at thirty seven degree centigrade (37 °C) for 15 min. The reaction was terminated with 1.0 ml DNSA reagent, placed in a boiling water bath for five min, then allowed to cool to room temperature and the absorbance was recorded at 540 nm. The control amylase exhibited 100% enzyme activity and did not contain any sample of analysis. To eliminate the absorbance by sample, the appropriate extract controls with the extract in the reaction mixture in which the enzyme was added after adding DNS (Fig. 1). The liberated sugar was then determined by the help of standard maltose curve and activities were calculated according to the following formula given below: Activity =
Conc.of Maltose libereted × ml of enzyme used × dillution factor Mol.wt of Maltose × incubation time (min) (1)
One unit of enzyme activity is defined as the amount of enzyme required to release one micromole of maltose from starch per min under the assay conditions. The inhibitory/induction property shown by the sample was compared with that of control and expressed as percent induction/inhibition. This was calculated according to the following formula: % inhibition/induction =
Activity in presence of compound × 100 Control Activity
(2)
2.5 Analysis of Acarbose as Standard Inhibitor Acarbose was used as a standard inhibitor and it was assayed at above-mentioned test sample concentrations. The assay method was similar to the above-mentioned procedure, instead of test samples, acarbose was added. The results were compared to that of test sample.
Comparative Evaluation of in Vitro Antioxidant …
133
Fig. 1 Schematic representation of preparation of nanoparticle formulations
2.6 Determination of Total Antioxidant Capacity Using a series of test tubes samples were taken in different concentrations (10, 50, and 100 μg). To this, 1.9 mL of reagent solution (0.6 M sulfuric acid, 28 mM sodium phosphate, and 4 mM ammonium molybdate) was added. Then the tubes were allowed to incubate at 95 °C for ninety (90) min after that allowed to cool. The absorbance of the solution of each was recorded at 695 nm against a blank.
134
T. Satapathy et al.
Antioxidant capacities were expressed as equivalents of ascorbic acid. Ascorbic acid equivalents were calculated using standard graph of ascorbic acid. Butylated hydroxyanisole (BHA) was used as reference standard. The values were expressed as ascorbic acid equivalents in μg per mg of extract (Prieto et al. 1999).
2.7 Bacterial Strain-Based Cytotoxicity Screening: The bacteria E.coli AB 1157 (wild-type strain), proficient to repair damage in the DNA was considered. Tryptone ten gm, NaCl ten gm, and Yeast extract five gm, Agar twenty gm in 1000 ml of distilled water was prepared and used as a media for this study. Initially, the stock culture of bacteria was revived by inoculating in broth medium and grown at 37 ºC for eighteen hrs. The LB Agar plates were prepared and wells were made in the solidified LB agar plate. Each plate was allowed to inoculate with eighteen hour old cultures (100 μl, 10–4 cfu) and spread evenly on the plate. After twenty min, the wells were filled with compound at different concentrations. Standard compound plate was also prepared in the similar manner. All the plates were incubated at 37 ºC for twenty hour and the diameter of inhibition zone were noted (Figs. 2, 3, 4, 5, 6, 7 and Tables 1, 2, 3).
3 Results and Discussion 3.1 Amylase Inhibition by Different Nanoformulation In the present study, the nanoparticles were successfully prepared by nano coprecipitation method. We have used chitosan as polymer and PCL as copolymer. All the formulations were spherical in shape. Different concentrations such as 100, 500, 1000 mg of three different formulations were screened for amylase inhibition assay. The results are represented in Table 1 and Fig 2. The % of amylase inhibition by three different formulations at 1000 μg concentrations were –404.92, –358.54, and –393.88 for curcumin nanoparticles, Cur + Pip dual drug-loaded nanoparticles and Piperine nanoparticles, respectively, but in case of total antioxidant activity, Cur + Pip dual drug-loaded nanoparticles at 100 μg showed better antioxidant activity i.e., 34.6 where other two formulations pip-NP and Cur-NP exhibited 30.5 and 28.2, respectively. This result for total antioxidant activity indicated that curcumin and piperine in combination possess better antioxidant activity than the other two formulations. In cytotoxicity study, none of the formulations exhibited any zone of inhibition hence no cytotoxic in nature.
Comparative Evaluation of in Vitro Antioxidant …
135
4 Conflict of Interest
Absorbance at 540 nm
The authors declare no conflict of interest.
1.4 y = 0.0121x + 0.0319 R² = 0.9975
1.2 1 0.8 0.6 0.4 0.2 0 0
20
40
60
80
100
120
Concentra on of Maltose (μg)
Fig. 3 Amylase inhibition assay by different nanoformulations
Ac vity (μmoles/ml/min
Fig. 2 Standard maltose curve
Amylase inhibition assay 0.08 0.06 0.04 0.02 0 Control
CUR NP
PIP NP
1000μg
500μg
100μg
Total antioxidant activity µg equvalent to Ascorbic acid
Fig. 4 Total antioxidant activity by different nanoformulations
CUR PIP NP
60 40 20 0
CUR NP CUR+PIP PIP NP
NP
10µg
50µg
BHA
100µg
136
Fig. 5 E.coli-based cytotoxic activity of Curcumin nanoparticles
Fig. 6 E.coli-based cytotoxic activity of Cur + Pip nanoparticles
T. Satapathy et al.
Comparative Evaluation of in Vitro Antioxidant …
137
Fig. 7 E.coli-based cytotoxic activity of Piperine nanoparticles
Table 1 Composition of different nano particle formulations Formulation
Curcumin(mg)
Piperine(mg)
Chitosan
PCL
Cur-NP
100
–
50
950
Pip-NP
–
100
50
950
Cur + Pip-NP
50
50
50
950
138
T. Satapathy et al.
Table 2 Effect of different nanoformulations on In vitro amylase inhibition Sample details
Concentrations
OD at 540 nm
Maltose liberated (μg)
Activity (μmoles/ml/min)
Activity (%)
Inhibition (%)
CUR-NP
Control
2.92
240.75
0.066817
536.00
−436.00
100 μg
2.4415
200.88
0.055751
447.22
−347.22
500 μg
2.5535
210.21
0.058341
468.00
−368.00
1000 μg
2.7525
226.79
0.062943
504.92
−404.92
100 μg
2.4675
203.04
0.056352
452.05
−352.05
500 μg
2.3545
193.63
0.053738
431.08
−331.08
1000 μg
2.5025
205.96
0.057161
458.54
−358.54
100 μg
2.311
190.00
0.052732
423.01
−323.01
500 μg
2.4585
202.29
0.056144
450.38
−350.38
1000 μg
2.693
221.83
0.061567
493.88
−393.88
CUR-PIP-NP
PIP-NP
Table 3 Effect of different nanoformulations on total antioxidant activity Concentration
Total antioxidant activity CUR-NP
CUR + PIP-NP
PIP-NP
BHA
10 μg
13.6
21.65
14
50 μg
21.05
25.4
23.5
29.9
100 μg
28.2
34.6
30.5
45.73
8.18
Acknowledgements Authors are thankful to Sunpure extract Pvt Ltd for providing Curcumin and Piperine as a gift sample for this research work.
References Abdel-Mageid AD, Abou-Salem MES, Salaam NMHA, El-Garhy HAS (2018) The potential effect of garlic extract and curcumin nanoparticles against complication accompanied with experimentally induced diabetes in rats. Phytomedicine 43:126–134 Aggarwal BB, Harikumar KB (2009) Potential therapeutic effects of Curcumin, the antiinflammatory agent, against neurodegenerative, cardiovascular, pulmonary, metabolic, autoimmune and neoplastic diseases. Int J Biochem Cell Biol 1:40–59 Aggarwal BB, Kumar A, Bharti AC (2003) Anticancer potential of curcumin: pre-clinical and Clinical studies. Anticancer Res 23:363–398 Anand P, Thomas SG, Kunnumakkara AB, Sundaram C, Harikumar KB, Sung B (2008) Biological activities of curcumin and its analogues (Congeners) made by man and mother nature. BiochemPharmacol 76:1590−1611 Belakavadi M, Salimath BP (2005) Mechanism of inhibition of ascites tumor growth in mice by curcumin is mediated by NF-kB and caspase activated Dnase. Mol Cell Biochem 273:57–67
Comparative Evaluation of in Vitro Antioxidant …
139
Gururaj AE, Belakavadi M, Venkatesh DA, Marme D, Salimath BP (2002) Molecular mechanisms of anti-angiogenic effect of curcumin. Biochem Biophys Res Commun 297:934–942 Jia T, Rao J, Zou L, Zhao S, Yi Z, Wu B, Li L, Yuan H, Shi L, Zhang C, Gao Y, Liu S, Xu H, Liu H, Liang S, Li G (2018) Nanoparticle-Encapsulated Curcumin Inhibits Diabetic Neuropathic Pain Involving the P2Y12 Receptor in the Dorsal Root Ganglia. 11:01–12 Joshi LS, Pawar AH (2015) Herbal cosmetics and cosmeceuticals: an overview. Nat Prod Chem Res 3:170 Miller GL (1959) Use of dinitrosalicylic acid reagent for determination of reducing sugar. Anal Chem 31:426–428 Prieto P, Pineda M, Aguilar M (1999) Spectrophotometric quantitation of antioxidant capacity through the formation of a phosphomolybdenum complex: specific application to the determination of vitamin E1. Anal Biochem 269:337–341 Shishodia S, Amin HM, Lai R, Aggarwal BB (2005) Curcumin (diferuloylmethane) inhibits constitutive NF-kappa B activation, induces G1/S arrest, suppresses proliferation, and induces apoptosis in mantle cell lymphoma. Biochem Pharmacol 70:700–713 Sudha P, Smita SZ, Shobha Y, Bhargava ARK (2011) Potent a-amylase inhibitory activity of Indian Ayurvedic medicinal plants. BMC complementary and alternative medicine 11:5 SundarDhilip Kumar S, Houreld N, Abrahamse H (2018) Therapeutic potential and recent advances of curcumin in the treatment of aging-associated diseases. Molecules 23(4):835
Improved ERP Classification Algorithm for Brain–Computer Interface of ALS Patient Vyom Raj, Shreya Sharma, Mridu Sahu, and Samrudhi Mohdiwale
Abstract The study on Amyotrophic Lateral Sclerosis (ALS) patient to identify the non-target or target stimulus based on event-related potential provide a way to improve P300 speller based Brain–Computer Interface (BCI). In the current work channel wise EEG data taken for the research. Feature extraction and Feature selection techniques based on Fourier and Wavelet transform and Statistics have been implemented to get the required features among the redundant one. By classifying the features categorized in 3 labels stated above by using support vector machine (SVM). The study reveals that the classification accuracy is improved in Morlet wavelet-based feature than statistical features for different channels taken in consideration. Keywords Brain computer interface · Amyotrophic lateral sclerosis · Wavelet ransform
1 Introduction Muscle movements such as speaking, walking, etc., are caused by motor neurons. The loss of functioning of motor neuron or decay of nerve cell causes a neurological disease named Amyotrophic Lateral Sclerosis (ALS). The person affected by this disease loses its control over the muscle movement due to death of motor neuron. The signal transmission process of muscle movement starts from brain’s upper motor neuron called cerebral cortex and move toward spinal cord, i.e., lower motor neuron, from the spinal cord the signal reached to specific muscle and perform functioning. As the process starts from upper motor neuron so, the degenerative disease brings gradual impairment starts from upper motor neuron to the spinal cord. ALS patient suffer from cognitive impairments, become quadriplegic and loses ability of communication even gestural communication due to complete failure of limb functioning (PietroCipresso et al. 2012; Mitsumoto and Rabkin 2007). Such critical cases attract the researchers to work on ALS patient to make their life easier. No physical movement needs special V. Raj · S. Sharma · M. Sahu (B) · S. Mohdiwale National Institute of Technology Raipur, Raipur, India e-mail: [email protected] © Springer Nature Singapore Pte Ltd. 2021 A. A. Rizvanov et al. (eds.), Advances in Biomedical Engineering and Technology, Lecture Notes in Bioengineering, https://doi.org/10.1007/978-981-15-6329-4_13
141
142
V. Raj et al.
care and support in all the aspect of life such as dietary, rehabilitation, respiration, and communication via devices. Brain–computer interface (BCI) device is the recent advancement for ALS patient to talk with their dear ones and communicate. The main concept of BCI is to transform the brain signal, i.e., EEG signal to respective word which enable a patient to talk via thinking. Deciphering the EEG data itself is a challenging task. Various methods are proposed to improve the performance of BCI. In contribution toward improvement of BCI we proposed a hybrid algorithm for feature extraction and selection. Performance evaluation of different wavelet has been done via Support Vector Machine (SVM). The rest of the paper structured as follows. Section 2 describes the literature review related to current work. Section 3 provides description of dataset. Section 4 provides the detail description of proposed method for performance improvement. Section 5 provides the result and analysis. Conclusion of the current work given in Sect. 6.
2 Literature Review Today’s era is digital era of communication, with this regard BCI plays very important role in communication for patient suffered from ALS disease. Nicolas F Ramsey et al. presented a fMRI-based method of BCI implant of new generation BCI (Ramsey 2014). The complete flow of BCI-based EEG signal processing shown in Fig. 1 Various researches is going for making the life easy of ALS patient based on EEG signal processing. Hao-Teng Hsu et al. explores amplitude-frequency characteristics of steady-state visual evoked potential (SSVEP) for obtaining the implementation feasibility of frontal SSVEP on BCI system. In this paper authors found frontal SSVEP on no hair-bearing area provides more comfort for BCI-based application due to lower amplitude on frontal SSVEP-based region than occipital SSVEP-based region (Hsu and I-Hui Lee 2016). Rosario Sorbello et al. improves the BCI performance by utilizing biofeedback factor obtained by considering the mental state of
Fig. 1 Complete work flow of BCI using EEG signal
Improved ERP Classification Algorithm …
143
human, futher it shows performance of the robotic feedback by post hoc analysis for maintaining the focus on expected task (Sorbello et al. 2018). The research is going on to evaluate the current potential for BCI-based P300 speller. This potential can be used further for online BCI-based applications (SadafIqbal et al. 2017). Yazıcı et al. proposed statistical feature-based EEG signal classification without preprocessing. The result obtained is higher efficient than the results uses pre-processing steps (Yazıcı and Mustafa Ulutas 2015). Faraz Akram et al. proposed an algorithm for word typing using P300 speller system with smart dictionary of suggesting word (FarazAkram and Metwally 2013). Yoon et al. investigates the spatial and temporal features of ERP and the result shows that the spatial features are more important than temporal features (Yoon et al. 2018).Thomas P. et al. presents spectral feature-based classification for biomedical signal. Multi wavelet transform, Independent component analysis are the extracted features for signal classification in that paper and found accuracy of 92% with SVM but on different dataset (Thomas and Moni 2016). Hong, K. Set al. presents the feature extraction and classification method for EEG-BCI, in which power spectral density, logarithmic band power are extracted features and linear discriminant analysis is most widely used classifier (Hong et al. 2018). Various literature discussed above motivate us to research in the area of BCI for improvement in the devices designed for ALS patient. In the current work, we focused on ERP classification using wavelet-based feature extraction and selection technique and find the effect of number of channels selected while evaluation of technique.
3 Dataset Description BCI2000 dataset has taken for the research in the current study. The dataset consists of recordings of P300 signals obtained from BCI2000 with the help of technique discussed by Farewell and Donchin. 8 patients suffering from ALS disease had to gaze one character out of 36 different character and the task is to predict the correct character while recording of EEG. The recordings are characterized in 3 different states, i.e., no stimulus, non-target stimulus, target stimulus.
4 Methodology The process of improving the feature extraction and selection method involve definite steps, such as Data acquisition, Feature extraction, Feature selection, and Classification. The Flowchart shown below represents the steps wise procedure followed for enhancement of existing techniques (Fig. 2). The details of each step provided in subsequent subsection.
144
V. Raj et al.
Fig. 2 Flowchart of proposed method
4.1 Data Acquisition BCI 2000 dataset is used for performing the research on ALS patient. The dataset has been obtained on 8 channel EEG acquisition device. To perform the experiment on each channel EEG data, channel wise data has been imported. Eight channels that were used for recording EEG based on the 10–10 standard (Fz, Cz, Pz, Oz, P3, P4, PO7, and PO8) (Riccio et al. 2013). The obtained EEG signal sampled at 256 Hz. Bandpass filter with band frequency 0.1 to 30 Hz has been used to get the desired frequency range of EEG signal.
4.2 Feature Extraction The large amount of data always may or may not be important. Introducing large amount of redundant data always maximizes the complexity of the analysis. Features are the important characteristics of data represented in less dimensionality than original data to identify data belongs to particular class. In the current work, features of EEG signal extracted by EEG Analysis package of R. Windowing of original data, double windowing, Spectrum of windowed data, Continuous wavelet transform are the feature extraction methods taken to obtain the features. Fast Fourier Transform. Fast Fourier Transform (FFT) is an efficient algorithm for finding the Discrete Fourier Transform (DFT) of a sequence with reduced time complexity. FFT divides the signal into small frames over a period of time and differentiate it according to frequency component present in the signal (Welch 1967). For N components, FFT can be calculated by using formula: N /2−1
X [k] =
n=0
N /2−1
x[2n]W N2kn + W N k
x[2n + 1]W N2 kn
(1)
n=0
where x(n) is signal of length n, W N kn is twiddle factor. FFT is calculated by dividing the signal into even and odd components. The above two terms in the Eq. 1 are those even and odd components of EEG signal.
Improved ERP Classification Algorithm …
145
Significance of calculating FFT for ALS patient data is to obtain the different frequency components of EEG data in discretized manner to evaluate them according to frequency bands of EEG signals. Wavelet Transform. Wavelet transform offers time–frequency representation of data/signal for localization, but not only in the sin or cosine, i.e., Fourier domain but also in real space. In the current work HAAR, Gaussian derivatives, and Morlet wavelet has been implemented for feature extraction from the dataset. The general function for calculating wavelet transform is given in Eq. 2 (SulimanBelal et al. 2018). f (a, b) =
∞
−∞
∗ f (x)ψa,b (x)d(x)
(2)
where ψ a,b (x) is wavelet function chosen. The HAAR wavelet is a square-shaped function which allows the target function to be represented in orthonormal basis. As this wavelet function is not continuous hence can-not be differentiable which allows to analyze the signals with sudden changes. Gausian and Morlet wavelets are less sensitive and allow trade between spatial and frequency domain resolution. Thus, they are more popular.
4.3 Feature Selection Acquiring relevant feature from the set of features for simplification, less training time with improved efficiency and generalization is the main objective of feature selection. In the current work False Discovery Rate (FDR), Analysis of Variance (ANOVA) has been implemented for feature selection problem. FDR provides the number of false observations that could be rejected and ANOVA provides the statistical test among different 8 channel’s data to analyze the differences among different channels. FDR is the method to adjust the p-value of test such that the total false positive errors are less (Benjamini and Hochberg 1995). For V number of false positives and R number of rejected null hypothesis FDR can be calculated by Z =
V R
F D R = E(Z )
(3) (4)
ANOVA is the technique used to analyze the differences among the mean of a group in a given sample of population. This is useful for comparing and choosing one sample from the set of samples (Sokal and Rohlf 1969). This value also useful to rank the features or variable which tells more discrimination among different classes.
146
V. Raj et al.
Fig. 3 Distribution of features after feature selection
In the current work, after feature selection a feature vector of length 304*5 has been obtained. After feature selection the distribution of feature shown in Fig. 3. After feature selection statistical operations are evaluated channel wise to get optimum feature. Mean, median, product, geometric and harmonic mean are calculated. On the same set of features obtained after selection procedure HAAR, Gaussian, Morlet wavelet is applied for another set of optimum features. These sets of features are now classified to check performance of proposed method.
4.4 Classification After accomplishment of data acquisition channel wise, feature extraction and feature selection, Support Vector Machine(SVM) is applied for classification of acquired features from different statistical and Wavelet Transforms. SVM transforms the data into high dimensional feature space and creates the decision boundary by maximizing the margin on obtained support vector. Various kernel functions can be used as per the data distribution for accurate classification. In the current work polynomial kernel is used as it is efficient as per data distribution.
5 Results In Table 1, classification accuracy after the application of SVM is presented. It is evident that the best classification accuracy is obtained when we have all the 8 channels taken into consideration and the conditional measure chosen is variance.
Improved ERP Classification Algorithm …
147
Table 1 Channel Wise Accuracy on Statistical Parameters Channel
Mean
Median
Variance
Product
Geometric
Harmonic
Fz
63.02
78.21
79.43
74.50
63.27
71.63
Fz,Cz
62.71
81.01
72.53
71.77
82.50
73.90
Fz,Cz,Pz
64.09
83.29
73.89
74.42
83.10
74.62
Fz,Cz,Pz,Oz
66.72
84.11
74.21
75.77
84.77
77.25
Fz,Cz,Pz,Oz,P3
71.41
85.22
77.23
75.91
86.55
78.21
Fz,Cz,Pz,Oz,P3,P4
76.83
87.11
88.92
77.64
81.2
89.27
Fz,Cz,Pz,Oz,P3,P4,PO7
82.42
85.11
95.33
78.91
82.34
84.50
Fz,Cz,Pz,Oz,P3,P4,PO7,PO8
88.92
89.11
97.83
80.91
90.61
93.24
Fig. 4 Conditional Statistics and Accuracy Channel Wise
When all 8 channels are chosen along with the conditional measure variance, we get the classification accuracy of 97.83%. The bar graph in figure also represent the channel wise conditional statistics and accuracy (Fig. 4). Further we went on to choose the best types of wavelet transformation that could increase classification accuracy. The results of the same have been mapped in Table 2. It has been established that the Morlet method of wavelet transformation gives the best classification accuracy. Combined with variance conditional measure it increases the classification accuracy to 98.28% (Fig. 5).
6 Conclusion The classification of dataset after feature extraction and selection while varying certain parameters have been accomplished in the current work. The main idea was to obtain the best conditional measures and the optimum number of channels. From
148
V. Raj et al.
Table 2 Channel wise accuracy on different wavelet transforms Channels
Haar
Gaussian1
Gaussian2
Morlet
Fz
80.41
82.93
80.43
72.42
Fz,Cz
74.52
76.21
77.64
77.61
Fz,Cz,Pz
75.62
76.54
70.33
82.45
Fz,Cz,Pz,Oz
76.51
77.89
75.31
83.89
Fz,Cz,Pz,Oz,P3
78.11
79.03
80.41
87.61
Fz,Cz,Pz,Oz,P3,P4
89.31
90.21
92.22
93.22
Fz,Cz,Pz,Oz,P3,P4,PO7
96.24
96.72
93.42
96.72
Fz,Cz,Pz,Oz,P3,P4,PO7,PO8
97.91
97.94
96.21
98.28
Fig. 5 Wavelet Transformation and Accuracy Channel Wise
the experiment it is clear that if the number of channels is incremented it increases the classification accuracy. On the other hand morlet wavelet transform also improves the performance of classification.
References Akram F, Metwally MK, Han H-S, Jeon H-J, Kim T-S (2013) A novel p300-based bci system for words typing. In Brain-Computer Interface (BCI), 2013 International Winter Workshop on, ppp 24–25. IEEE Belal S, Cousins J, El-Deredy W, Parkes L, Schneider J, Tsujimura H, Zoumpoulaki A, Perapoch M, Santamaria L, Lewis P (2018) Identification of memory reactivation during sleep by eeg classification. NeuroImage 176:203–214
Improved ERP Classification Algorithm …
149
Benjamini Y, Hochberg Y (1995) Controlling the false discovery rate: a practical and powerful approach to multiple testing. J Royal Statist Soc Series B (Methodological), 289–300 Cipresso P, Carelli L, Solca F, Meazzi D, Meriggi P, Poletti B, Lul´ e D, Ludolph AC, Silani V, Riva G (2012) The use of p300-based bcis´ in amyotrophic lateral sclerosis: from augmentative and alternative communication to cognitive assessment. Brain and behavior 2(4):479–498 Hong KS, Khan MJ, Hong MJ (2018) Feature extraction and classification methods for hybrid fNIRS-EEG brain-computer interfaces. Frontiers in Human Neuroscience, 12 Hsu H-T, Lee I-H, Tsai H-T, Chang H-C, Shyu K-K, Hsu C-C, Chang H-H, Yeh T-K, Chang C-Y, Lee P-L (2016) Evaluate the feasibility of using frontal ssvep to implement an ssvep-based bci in young, elderly and als groups. IEEE Trans Neural Sys Rehabil Eng 24(5):603–615 Iqbal S, Rizvi BA, Muhammed Shanir PP, Khan YU, Farooq O (2017) Detecting p300 potential for speller bci. In: Communication and Signal Processing (ICCSP), 2017 International Conference on, pp 0295–0298. IEEE. Mitsumoto H, Rabkin JG (2007) Palliative care for patients with amyotrophic lateral sclerosis:prepare for the worst and hope for the best. Jama 298(2):207–216 Ramsey NF (2014) Exploration of the brain for optimal placement of bci implants in paralyzed people. In Brain-Computer Interface (BCI), 2014 International Winter Workshop on, pp 1–3. IEEE Riccio A, Simione L, Schettini F, Pizzimenti A, Inghilleri M, Belardinelli MO, Mattia D, Cincotti F (2013) Attention and p300-based bci performance in people with amyotrophic lateral sclerosis. Frontiers in human neuroscience 7:732 Sorbello R, Tramonte S, Emanuele Giardina M, La Bella V, Spataro R, Allison B, Guger C, Chella A (2018) A human–humanoid interaction through the use of bci for locked-in als patients using neuro-biological feedback fusion. IEEE Trans Neural Sys Rehabil Eng 26(2):487–497 Sokal RR, Rohlf FJ (1969) The principles and practice of statistics in biological research. WH Freeman and company, San Francisco, pp 399–400 Thomas P, Moni R (2016) Methods for improving the classification accuracy of biomedical signals based on spectral features. Technology 7(1):105–116 Welch P (1967) The use of fast fourier transform for the estimation of power spectra: a method based on time averaging over short, modified periodograms. IEEE Trans Audio Electroacoust 15(2):70–73 Yazıcı M, Ulutas M (2015) Classification of eeg signals using time domain features. In 2015 23nd Signal Processing and Communications Applications Conference (SIU) pp 2358–2361 Yoon J, Lee J, Whang M (2018) Spatial and time domain feature of ERP speller system extracted via convolutional neural network. Computational intelligence and neuroscience
Size Reduction in Multiband Planar Antenna for Wireless Applications Using Current Distribution Technique Pravin Tajane and P. L. Zade
Abstract The proposed planar antenna with U shaped slots on patch and asymmetrically rectangular slots on ground plane for WLAN/Bluetooth/WiMAX/HYPERLAN applications. The proposed antenna is reduced the size by using the surface current distribution technique for the multiband applications. The surface current shows that which part is responsible to create the resonance frequencies. The maximum current is concentrated in the rectangular slots for 2.42, 3.41, and 5.4 GHz frequencies. Those portions do not have any surface current that part is not responsible to produce the resonance frequencies. The designed antenna 1 doesn’t have any surface current at the upper side of patch means that part does not take part actively participation to create the resonance frequencies so those portion is removed without affecting anything on other parameter. The planar antenna 2 is applicable for multiband applications after size reduction in antenna 1 using surface current distribution technique. The Proposed antenna with U shaped slot on patch and asymmetrically rectangular slots on ground plane which resonates 2.42 GHz, resonates 3.41 and 5.4 GHz covering ISM (Industrial, Scietific, and Medical) and WLAN, Bluetooth, Zigbee, WiMAX, and HYPERLAN. The length is reduced 40 to 30 mm with the help of current spread technique. An overall dimension of antenna is 30 × 26mm2 from 40 × 26 mm2 . After reducing the size of antenna is also applicable for the multiband operation. Keywords Multiband · Surface current · Rectangular slots
P. Tajane (B) YCCE, Nagpur, India e-mail: [email protected] P. L. Zade DMIETR, Wardha, India e-mail: [email protected] © Springer Nature Singapore Pte Ltd. 2021 A. A. Rizvanov et al. (eds.), Advances in Biomedical Engineering and Technology, Lecture Notes in Bioengineering, https://doi.org/10.1007/978-981-15-6329-4_14
151
152
P. Tajane and P. L. Zade
1 Introduction 1.1 Background Now a day’s microstrip patch antenna became very popular due to low size, easily fabricated, low weight, portable, and easy to install with electronic circuits. So the patch antenna is easily embedded with any type of devices which may portable or non portable. There are different band in different wireless application which is the alternative of cable so we can easily transmit the data from source to destination without disturbing of any information. There are many wireless applications such as industrial, medical, and scientific application (ISM band) use in various portable or non-portable devices for cordless head phone, baby monitoring, and Bluetooth earpiece, etc. Earlier antenna is used the single band for single application by using simple formulae. From the last decade instead of using single band for single antenna, the multiband antenna became very popular by using slits, slots, and partial ground plane technique. The size of antenna is reduced by taking the help of slits, slots, surface current distribution, and defected ground structure. The multiband antenna is applicable to different band in several portable advance devices such as laptop, cellular phone sets, stylish phones, etc. A multiple of antenna designs such as series capacitor is presented to reduce the size of antenna and double circular slot ring resonator is produced multiple band (Singh and Mahesh 2018), the square shape slot at ground plane reduce size and produce multiband (Melkeri and Hunagund 2017), combination of E and T shape slots on the patch and some ground plane adjustment is responsible to shifted frequency 3 to 0.9 GHz which reduce the size of size of antenna (Ambh and Singhal 2016), Couple of circular rings is added in ground plane to reduce the size of antenna for maintaining reference band (Lastname et al. 2015), the pair of mirrored L-shaped strips for multiband operation is used in Tshaped monopole antenna which reduce size of antenna (Jui-Han 2015), Meander slots is responsible to reduce the size of antenna (Brocker et al. 2014), Spiral planar inverted F is used to reduce the size of patch antenna (Achmad Munir 2014), the defected microstrip structure is used to the minimize size of antenna (Singh 2011), Koch shaped fractal defect is used in patch antenna to reduce the size of antenna (Kordzadeh and Hojat Kashani 2009) instead of Split-Ring Resonator, the Complementary SplitRing Resonator (CSRR) is used to reduce the size of antenna (karimzadeh Baeel et al. 2007), G strip on front side and U strips on back side of substrate is used to reduce the size of antenna as well as to bring several band (Prabhu et al. 2018), circular cross slots are used on patch and ground plane to reduce the size of antenna (Nada and Tawfeeq 2017), have been proposed different technique to reduce the size of antenna for several applications like GSM, WiMAX, WLAN, Bluetooth, Zigbeeand Hyperlan, etc.
Size Reduction in Multiband Planar Antenna …
153
1.2 Proposed Technique Easily we can designed single band by using predefined formulae for the different individuals application like WLAN, WiMAX, HYPERLAN, Zigbee, and Bluetooth. The same single band antenna is designed to produce multiple bands with the help of fractal shape, circular ring, slits, slots, metamaterials, etc. The size reduction of antenna play a very important role in portable devices such as cellular phone. The multiband and size reduction of antenna is very important instead of using single band. The proposed antenna has surface current distribution technique to reduce the size of antenna by maintaining the multiband operations. The Surface current distribution technique is one good technique to reduce the size of antenna. In which area have more surface current and which area does not have surface current. So those portions of antenna do not covers with the surface current meaning of that part does not take in activity to radiate the electromagnetic wave. We can remove that part to reduce the size of antenna. The planar monopole antennas have less gain due to the small size of antenna because the electrical length of patch antenna is directly proportional to gain of antenna. The current is carried from port to end of copper part on patch or ground plane but disturbance is created in between end of copper part on patch or ground plane so the length of electric current is increased which improve the gain of antenna. The capacitive and inductive effect is responsible for the shifting center frequency from higher to lower side or lower to higher side. Fc = 1/2 π
√
LC L-inductance.
Fc-center frequency C-capacitance.
The Bandwidth enhancements are inversely related to quality factor of antenna. The quality factor depends on copper size of antenna. The bandwidth can be improved by adding slot in copper layer so copper area decreases means quality factor decreases. Bandwidth 1/Q.F. Q.F.-Quality factor.
The main aim to reduce the size of antenna with the help of surface current distribution on the back and front side of antenna. After reducing the size of proposed antenna is also worked industrial, scientific, and medical (ISM) applications. It can be assist the upcoming designers to reduce the size of antenna using the surface current distribution technique.
2 Antenna Design and Simulation Approach Initially designed of planar antenna is shown in Fig. 1 without size reduction. The FR4 substrate is used to designed antenna with relative permittivity 4.3 and height 1.6 mm.
154
P. Tajane and P. L. Zade
Fig. 1 Antenna 1
Figure 1 shows top view and back view of planar antenna which size is 40 × 26 mm2 . Figure 2a shows that where the maximum current is concentrated so that part is responsible to produce 2.41 GHz resonance frequency. Figure 2b shows that where the maximum current is concentred so that part is responsible to produce 3.5 GHz resonance frequency. Figure 2c shows that where the maximum is current concentrated so that part is responsible to produce 5.33 GHz resonance frequency.
Fig. 2 Surface current distribution for a 2.41 GHz, b 3.5 GHz, and c 5.3 GHz
Size Reduction in Multiband Planar Antenna …
155
Figure 3 shows presented antenna after reduction with the help of surface current distribution on the planar antenna. Antenna 1 is 40 mm length but above 30 mm to 40 mm that part does not responsible to produce 2.41, 3.5, and 5.3 GHz center frequencies. So those portions removed from the antenna 1 and reduced size of antenna 2 is 30 × 26 mm2 by maintaining the same resonance frequencies with useful bandwidth. There is 25% size reduction by using surface current distribution technique.
Fig. 3 Geometry of the proposed antenna after reduction (Antenna 2)
Fig. 4 Surface current distribution for the frequencies a 2.45 GHz, b 3.4 GHz, and c 5.422 GHz
156 Table 1 Antenna designed parameters with dimensions (units in mm)
P. Tajane and P. L. Zade Designing parameters
Dimensions (mm)
Designing parameters
Dimensions (mm)
W1
3
Wg3
4.5
W2
11.5
Wg4
3
W3
6
Wg5
2.5
W4
3
Wg6
2.5
L1
30
Wg7
2
L2
12
Wg8
1.5
L3
16.5
Lg1
12
L4
13
Lg2
5.25
Wg1
26
Lg3
3.45
Wg2
8
Lg4
2
The dimensions are given in Table 1 to designed the multiband planar monopole antenna which is helpful to reduce the size of antenna (Fig. 4).
3 Simulation Result The designed antenna 1 dimensions is 40 × 26 × 1.6 mm3 which is applicable for the multiband operations for the different resonance frequencies such as 2.42, 3.5, and 5.43 GHz shown in Fig. 5. By using the surface current distribution technique the
Fig. 5 Return Loss characteristics of designed antenna 1 and antenna 2
Size Reduction in Multiband Planar Antenna …
157
size is reduced in designed antenna 1 because those part does not having any surface current i.e. that portion does not take part to produce radiations shown in Fig. 2 for the resonance frequencies 2.42, 3.5, and 5.43 GHz. The dimensions of designed antenna 2 is 30 × 26 × 1.6 mm3 which is reduced from designed antenna 1 using surface current distribution technique. There is 25% size reduction without disturbing required bands which is present in designed antenna 1. The standard frequency range of Zigbee, Bluetooth, WLAN, and ISM is 2.4 to 2.5 GHz. The proposed antenna is covered from 2.39 to 2.51 GHz with bandwidth of 120 MHz. The standard frequency of WiMAX is 3.4 to 3.6 GHz range which is covered in proposed antenna with 3.2 to 3.63 GHz with bandwidth of 430 MHz. The HYPERLAN 1 AND HYPERLAN 2 is covered from 5.15 to 5.85 GHz frequency range but in presented antenna is covered from 5.12 to 5.91 GHz with bandwidth of 490 MHz. The proposed antenna is useful to required bandwidth for multiband operation. All above simulation is done with the help of computer simulation technology (CST) software. CST is electromagnetic simulator software which gives the behavior of electric and magnetic of designed antenna in time or frequency domain (Fig. 6).
4 Experimental Results The Vector Network Analyser (VNA) is used to find the results of fabricated planar monopole antenna. The actual physical view of fabricated planar monopole antenna is as shown in Fig. 7 (Table 2). In Fig. 7 shows the SMA connector is used to interface between fabricated antenna and Vector Network Analyser to find out the measure results of planar antenna. Figure 8 shows the measured results of proposed antenna for the reflection coefficient which is nearly same simulated results of designed antenna so the proposed planar antenna is applicable to the multiband oprations.
5 Conclusions In this paper, the size of planar monopole antenna is reduced by using surface current distribution technique without disturbing any required bands. The portion of antenna which does not have surface current that part does not radiates the electromagnetic waves at outward side. Whereas the portion having more current are responsible to produced the resonance frequencies. The simulated results of antenna 1 shows that, it is applicable to multiband operation by achieving required band with resonance frequencies shown in Fig. 5. After size reduction in antenna 1 using surface current distribution technique is also applicable for multiband operation shown in Figs. 5 and 8. The fabricated result of proposed antenna is nearly same with simulated results. This surface current distribution technique is very easy method to reduce the size of
158
P. Tajane and P. L. Zade
Fig. 6 Pattern of polar type radiation for a 2.54 GHz b 3.57 GHz c 5.43 GHz
antenna because we can easily see which part of antenna have much more surface current or which portion does not have surface current. Hence this technique is helpfulto reduce the size of antenna without affecting anything forwireless applications to required resonance frequencies. The ISM band ranges 2.4–2.5 GHz which is internationally reserved for industrial, scientific, and medical purpose excluding telecommunications. It is unlicensed band which do not require permission from government authorities. This band is useful in medical field like diathermy machines and hyperthermia therapy. These devices are created electromagnetic radiation which disturbs communication by using matching frequency. The diathermy machines use electromagnetic waves in the ISM bands to concern profound heating to body for recreation and curing. The hyperthermia therapy is used microwaves to warm tissue to destroy cancer cells.
Size Reduction in Multiband Planar Antenna …
Front side Fig. 7 Fabricated of the proposed planar antenna
Fig. 8 Measured Characteristics of S11
159
Bottom side
160
P. Tajane and P. L. Zade
Table 2 Comparison with proposed antenna and other authors References
Designed technique
Singh et al. (2018)
Modified double circular slot ring 68.83 resonator(MDCSRR) on ground plane and inter digital capacitor(IDC) on patch
Size reduction in %
Melkeri and Hunagund (2017) Square shaped slot at the ground plane
90
Ambh and Singhal (2016)
E and T shaped slots on ground plane
86.73
Ripin et al. (2015)
Circular ring on ground plane
42
Lu et al. (2015)
pair of mirrored L-shaped strips
20
Proposed
Surface current distribution technique
25
References Ambh A, Singhal PK (2016) Size reduction of rectangular microstrip patch antenna for gsm application. IRJET 3:255–258 Brocker DE, Werner DH, Werner PL (2014) Dual-band shorted patch antenna with significant size reduction using a meander slot. IEEE Antennas and Propagation Society International Symposium (APSURSI), USA karimzadeh Baeel R, Dadashzadehl G, Geran Kharakhilil F (2007) Using of CSRR and its equivalent circuit model in size reduction of microstrip antenna. IEEE Kordzadeh A, HojatKashani F (2009) A new reduced size microstrip patch antenna with fractal shaped defects. Prog Electromagnet Res 11:29–37 Kushwah VS, Tomar GS (2011) Size reduction of microstrip patch antenna using defected microstrip structures. International Conference on Communication Systems and Network Technologies, IEEE Lu J-H, Zeng B-R, Li Y-H (2015) Planar multi-band monopole antenna for WLAN/WiMAX applications. ISANP, conference January, Taiwan Melkeri VS, Hunagund PV (2017) Study and analysis of sqmsa with square shaped defected ground structure. IAIM IEEE, India Munir A, Harish A, Chairunnisa C (2014) Size reduction of UHF planar inverted-F antenna with patch geometry modification. IEEE, ISAP December Prabhu P, Manikandaswamy S, Saminathan T, Muthu Kumaran B (2018) A compact G and U strip folded planar multiband antenna for wireless applications. Int J Pure Appl Mathemat 118:77–83 Ripin N et.al. (2015) Size miniaturization and bandwidth enhancement in microstrip antenna on a couple circular rings Dgs. Int J Latest Res Sci Technol ISSN (Online): 2278–5299 4(4):27–30, July-August Singh AK, Abegaonkar MP, Koul SK (2018) Miniaturized multiband microstrip patch antenna using metamaterial loading for wireless application. Progress Electromagnet Res C, 83:71–82 Tawfeeq NN (2017) Size reduction and gain enhancement of a microstrip antenna using partially defected ground structure and circular/cross slots. Int J Elect Comput Eng 7(2):894~898, April 2017, Iraq
Classification of Hepatic Disease Using Machine Learning Algorithms Lokesh Singh, Rekh Ram Janghel, and Satya Prakash Sahu
Abstract Liver is one of the biggest gland and only organ of the human body which can restore harmed cells. Disorders in liver function affects food digestion and releasing of harmful toxic substances from the body which may results in severe liver diseases like jaundice, abdominal pain abdominal swelling, etc. These diseases require clinical care by experts. Early diagnosis is the only demand that remains to prevent this speedy loss of liver function. Only few of the specific algorithms are employed to operate medical instruments (CT-scan, MRI, EEG, ECG, etc.) to diminish time and the expenditure in diagnosing the liver disease. This work demonstrates the use of numerous classification methods (which were not experimented over ILPD dataset) in designing the classification model. The performance of classification methods are assessed over several measures. ILPD dataset sourced from UCI machine learning repository is used in performing the experiments. Nine classifiers of two classification models are employed in experimentation, namely, SVM and Ensemble. Classifiers are analyzed and compared on various validation measures. Results prove that Coarse Gaussian SVM achieves highest accuracy of 71.4% with less training time achieved of 0.61 s. Keywords SVM · Ensemble · Coarse gaussian SVM · Machine learning · ILPD
L. Singh (B) · R. R. Janghel · S. P. Sahu National Institute of Technology, G.E Road, Raipur 492001, India e-mail: [email protected] R. R. Janghel e-mail: [email protected] S. P. Sahu e-mail: [email protected] © Springer Nature Singapore Pte Ltd. 2021 A. A. Rizvanov et al. (eds.), Advances in Biomedical Engineering and Technology, Lecture Notes in Bioengineering, https://doi.org/10.1007/978-981-15-6329-4_15
161
162
L. Singh et al.
1 Introduction Liver is the biggest meaty organ of the body which fulfills various vital functions including detoxification and releasing of harmful chemicals from the body increases the metabolism. Among its various functions, the main function of liver is the blood purification before allowing it to flow in rest of parts of the body. It is helpful in blood clotting by producing proteins (Vijayarani and Dhayanand 2015). In the field of computer science machine learning classification algorithms are appropriate for the diagnosis of various diseases and can be employed for designing automated medical diagnosis device. There may be certain factors like obesity, intake of drugs on regular basis, mental stress, consumption of alcohol, contaminated water, etc., which may be the cause of liver disorders like liver cancer, hepatitis—A, B, C, D, and E, etc. Indications of liver disease includes—abdominal ache and swelling, intensify itching, nausea, exhaustion, back-ache, etc. These and many more symptoms of liver disorders are hard to diagnose at an early stage as liver continues functioning when it is partially injured. Early prediction at an early stage is the only remedy to reduce the mortality rate due to liver disorders (Ghosh and Waheed 2017). Factors Affecting Lever (Rajeswari 2010) (Vijayarani and Dhayanand 2015): Various factors that might affect vital functions of liver if not focused like: Obesity, Viruses, Excessive use of alcohol, Hereditary, and Diabetes. Liver Disorders (Priya 2018): Common Liver disorders are discussed in this section. 1. Hepatitis: Lever usually gets infected by viruses like hepatitis A, B, and, C. There are some other causes of Hepatitis due to excessive intake of alcohol, medicines, obesity, etc. 2. Cirrhosis: Any factor which harms the liver can lead to Cirrhosis which makes the liver unfit to function in a proper manner. 3. Liver cancer: Cirrhosis may lead to lever cancer like Hepatocellular Carcinoma. 4. Liver failure: Several factors are responsible for the failure of liver including contaminated food, excessive usage of alcohol, obesity, and many more. 5. Hemochromatosis: one of the disorder responsible for damage of liver is the iron which silently sets in the liver due to hemochromatosis which leads to multiple health problems. Symptoms of Liver Disease (Gregorio Maldini): The following are the symptoms which sign liver disease are—Severe Jaundice, Intensify abdominal pain, Abnormal swelling, More confusion, Acute bleeding, Nausea, Pale stools, Increase in fatigue, and Sudden weight loss. Building Blocks of Liver: One of the major building block of liver is Phosphatidylcholine (PC) (Kidd 1996). Other building blocks are: 1. Vitamin B12—to manage nerve cells 2. Folic Acid—synthesis of DNA, RNA 3. Iron—for generation of red blood cells,
Classification of Hepatic Disease Using Machine Learning Algorithms
163
4. Vitamin A—for healthy vision, teeth 5. Vitamin D—promote bone growth 6. Vitamin K—blood clotting. The main purpose of designing this prediction model is the early diagnosis of liver diseases using machine learning classification algorithms which were never been implemented for classification of liver patient which may play an important role in reducing the mortality rate due to liver disease disorders. The results achieved are hopefully fruitful comparatively previously used algorithms in terms of various measures like accuracy and training time, etc. The rest of the research work is organized in following manner: Section-II describes Related Work regarding Liver Disease Prediction, Section-III defines the System Architecture, Classification Algorithms and Dataset used, Section-IV determines evaluated Results, and Section-V concludes the results of the research work through Conclusion.
2 Related Work Dinu et al. in (2017) focused on A.I-based techniques in designing a system for predicting and diagnosing various diseases which can help doctors or experts in analyzing the disease correctly. The strength of deep learning in designing such a system can raise up the true positive rate while reducing the false positive rates and improves the decision-making ability. The author has proved in this research that the employed A.I algorithm gives significant accuracy in diagnosing diseases. Details on Liver Disorder Diagnosis using Artificial Neural Network can also be found in (Janghel 2016). Ghosh et al. in (2017) performed experiments over UCLA and AP dataset (statistics course dataset). Several machine learning algorithms like K* algorithm, NBC, Bagging, Logistic, and Rep Tree are employed for diagnosing of diseases. All these algorithms are evaluated on various measures like accuracy, precision, sensitivity, and specificity. Results from the experiments show that K* algorithm outperforms comparatively other algorithms as it provides highest accuracy with minimum error rate and thus can diagnose liver disorders accurately. Kumar et al. in (2017) employed various machine learning classification algorithms like C4.5, Random Forest, CART, Random Tree, and REP tree for prediction of liver diseases. 80% of data is employed as training data and 20% of data is used for testing. The database results obtained from the experiment shows the highest accuracy of 79.22% from Random Forest algorithm. Eshraghi et al. in (2013) designed a model for the prediction of liver diseases. Eleven classification algorithms are analyzed and compared over the dataset on different factors like accuracy, precision, and recall. After performing various investigations Bayesian boosting is used in the experiment to achieve better results. Vijayarani et al. in (2015) used two machine learning algorithms named Naïve Bayes and Support Vector Machine for diagnosing liver disease. These two approaches are analyzed and compared over accuracy and execution time
164
L. Singh et al.
Fig. 1 Proposed model diagram
measures. During experimentation SVM has achieved highest accuracy while Naïve Bayes has achieved minimum execution time.
3 Classification Experiments 3.1 System Architecture Proposed model diagram is depicted in Fig. 1. The diagram shows the process undergone in the research.
3.2 Normalization Normalization can be defined as a scaling approach or can be referred as mapping method or a preprocessing stage. Through Normalization we can obtain novel range from the current range. It tends to be useful in approaches like prediction (Patro and Kumar).
3.3 Classification Algorithms For experimentation, classification models are categorized into two parts, namely, SVM and Ensemble, which comprises of various classifiers. Below is the brief description of algorithms used. Support Vector Machine. Support Vector Machine is a supervised learning approach firstly introduced by Cortes and Vapnik in 1995. Their objective behind designing the SVM is transforming binary class data into feature space or higher dimensional space and find an optimum separating hyperplane among set of hyperplanes which provides maximum margin between two different classes (Sisodia 2010). SVM is a binary classifier which has the ability to classify both linear and non-linear separable
Classification of Hepatic Disease Using Machine Learning Algorithms
165
data. SVM locates the decision boundary by very small subset of training samples called as the support vectors (Sisodia 2014). The class y is defined as follows: y
(i)
=
−1 i f w T x (i) + b ≤ −1 1 i f w T x (i) + b ≤ 1
(1)
Can also be written as y(i) (w T x (i) + b) ≥ 1. The objective of SVM is to fulfill two requirements: a. There should be maximum possible distance between the two decision boundaries. The equation of hyperplane can be defined as wT x + b = 1 and the distance between the hyperplane can be defined mathematically as wT x + b = 1 which 2 . If we should be maximize. The distance between the margin is defined as ||w|| 2 want to maximize the distance i.e. max ||w|| , equivalently we have to minimize
min ||w|| . 2 b. Support Vector Machine should also correctly classify all X(i) , which is the ith sample in the dataset, i.e.,
y (i) (w T x (i) + b) ≥ 1,
∀i ∈ {1, ...., N }
(2)
Linear SVM, Cubic SVM, Quadratic SVM, Fine Gaussian, Medium Gaussian SVM, and Coarse Gaussian SVM are the classifiers employed in SVM classification model (Alpaydin 2010). Comparative description of different SVM Classifiers are described in Table 1 [https://www.mathworks.com/help/stats/choose-a-classifier. html]. Ensemble. In supervised learning Ensemble classifiers are defined as set of individual classifiers which are trained on a database in a supervised fashion. The concept behind the ensemble classifier is the recompense of errors (Rahman and Tasnim 2014). It works in three steps: in first step it creates multiple datasets from the original training data, in second step it multiple classifiers, in third step it combines all the classifiers to obtain more accurate predictions using weighted majority voting. Three methodologies are employed in designing an effective ensemble system. Following three methodologies are known as basic pillars of ensembles. (1) Sampling of data, (2) training of member classifiers, and (3) combining the classifiers. Ensemble methods not only tends to be useful in reducing the variance but also helpful in increasing the generality very effectively. Ensemble-based systems gain their strength in managing the large volume of data in an incremental manner (Cha Zhang and Yunqian Ma 2012) Boosted Trees, Bagged Trees, and RUS Boosted Trees are the classifiers employed in Ensemble classification model. Table 2 represents comparative description of all ensemble classifiers [https://www.mathworks.com/help/stats/choose-a-classifier. html].
166
L. Singh et al.
Table 1 Comparative description of all SVM classifiers Classifier type
Prediction speed
Memory usage
Interpretability
Model flexibility
Linear SVM
Binary: Fast Multiclass: Medium
Medium
Easy
Low, Makes a simple linear separation between classes
Quadratic SVM
Binary: Fast Multiclass: Medium
Binary: Medium Multiclass: Large
Hard
Medium
Cubic SVM
Binary: Fast Multiclass: Medium
Binary: Medium Multiclass: Large
Hard
Medium
Fine Gaussian SVM
Binary: Fast Multiclass: Medium
Binary: Medium Multiclass: Large
Hard
High, decreases with kernel scale setting. Makes finely detailed distinctions between classes, with kernel scale set to sqrt (P)/4
Medium Gaussian SVM
Binary: Fast Multiclass: Medium
Binary: Medium Multiclass: Large
Hard
Medium, Medium distinctions, with kernel scale set to sqrt (P)
Coarse Gaussian SVM
Binary: Fast Multiclass: Medium
Binary: Medium Multiclass: Large
Hard
Low, Makes coarse distinctions between classes, with kernel scale set to sqrt (P)*4, where P is the number of Predictors
Table 2 Comparative description of ensemble classifiers Classifier type
Prediction speed
Memory usage
Interpretability
Model flexibility
Boosted trees
Fast
Low
Hard
Medium to High
Bagged trees
Medium
High
Hard
High
RUS boosted trees
Fast
Low
Hard
Medium
Classification of Hepatic Disease Using Machine Learning Algorithms Table 3 Attribute information of the dataset
Description
167
S.No
Features
Type
1
Age
Age of the patient
Real number
2
Gender
Gender of the patient
Categorical
3
TB
Total bilirubin
Real number
4
DB
Direct bilirubin
Real number
5
Alkphos
Alkaline phosphotase
Integer
6
Sgpt
Alanine aminotransferase
Integer
7
Sgot
Aspartate aminotransferase
Integer
8
TP
Total protiens
Real number
9
ALB
Albumin
Real number
10
A/G Ratio
Albumin and globulin ratio
Real number
11
Class
Selector field used as class label
Binomial class
3.4 Dataset Used For experimental purpose MATLAB tool is used to perform the experiment. In this study ILPD (Indian Lever Patient Dataset) is chosen for building and testing classification models, which is taken from UCI repository. Table 3 shows attribute information of the dataset. The information in the dataset is collected from north east of Andhra Pradesh, India. The dataset comprises of 10 attributes and 1 selector field (class) which classifies the records as liver patient and non-liver patient. The dataset has a total 583 instances, out of which 416 records classifies liver patient and 167 records classifies non-liver patient. Out of all 583 instances 441 records are male patient’s records and 142 records are female patient records (Bendi Venkata Ramana, M. Surendra Prasad Babu 2018). Brief Description Of Attributes (Sumedh Sontakke, Jay Lohokare 2017): 1. 2. 3. 4. 5. 6.
Age: This attribute represents age of the patient. Gender: This attribute represents gender of the patient. TB: Bilirubin is a substance which is constructed during the break down of old red blood cells. It helps the liver in digesting the food. DB: Also known as conjugated bilirubin, it is usually passed in small amount through kidneys and excreted via urine. Alkphos: It is a type of enzyme found in our body in various tissues. It is used for the diagnosis of damage of liver and disorders in bone. Alanine Aminotransferase: Alanine Aminotransferase is a kind of enzyme found in the cells of our liver and kidney. It is considered as the initial screening test for diagnosing liver disease.
168
L. Singh et al.
Fig. 2 Scatter Plot of original dataset: Indian Patient Liver Dataset (ILPD)
7.
Aspartate Aminotransferase: It is a type of enzyme which when released shows the indication of damage of liver or muscles. 8. Total Protiens: Total protein test measures the total amount of albumin and globulin in our body. 9. Albumin: It is one of the main protein induced in the lever to maintain intravascular colloid osmotic pressure (COP). 10. A/G Ratio: The albumin/globulin (A/G) ratio is used to evaluate liver and kidney disease. 11. Class: It is a selector field used for binary classification. Scatter plot diagram of ILPD original dataset is shown in Fig. 2. Validation Metrics. Following are the details of the validation measures used in the experimentation. • Accuracy: Statistical measure of correctly predicted instances to the total no of instances. It can be defined as: A=
TP +TN T otal no o f samples
• Precision: Statistical measure of correctly predicted positive instances to the total no of predicted positive instances. It can be defined as: P=
TP T P + FP
Classification of Hepatic Disease Using Machine Learning Algorithms
169
• Recall: Measure of proportion of retrieved relevant observation among the total relevant observations. It can be defined as: R=
TP T P + FN
• F-1 Score: Measure of the weighted average of Precision and Recall. It can be defined as: F1 =
2T P (2T P + F P + F N )
• ROC: Receiver Operating Curve is Measure to plot true positive rate against the false-positive rate.
4 Results This section illustrates the results obtained in the experimentation using various classification methods which are deemed as superior data mining approaches employed in health-care for diagnosing the diseases. MATLAB tool is being used in experimentation. In total nine machine learning classifiers are employed in performing the experiment over ILPD dataset. tenfold cross-validation method is used for the validation of different classifiers. Table 4 represents comparative performance of different classifiers calculated over measures like Sensitivity, Specificity, Precision, and F-1 score. As seen from Table 4 there is substantial difference between the sensitivity, specificity, precision, and F1-score of SVM and ensemble methods. Out of six SVM classifiers highest sensitivity is achieved by Cubic SVM 77.52% while in ensemble classifiers the highest sensitivity of 88.01% is achieved by RUS Boosted Trees. Sensitivity is deemed as the diagnostic ability where a person is diagnosed with disease Table 4 Classification results of classifiers on various measures Model Type Classifiers
Sensitivity % Specificity % Precision % F1-Score %
SVM
Linear SVM
71.36
0.0
Quadratic SVM
75.20
48.42
88.22
81.19
Cubic SVM
77.52
45.51
79.57
78.53
Fine gaussian SVM
73.35
48.15
93.27
82.12
Ensemble
100.0
83.28
Medium gaussian SVM 71.36
0.0
100.0
83.28
Coarse gaussian SVM
71.36
0.0
100.0
83.28
Boosted trees
73.96
38.10
81.25
77.43
Bagged trees
75.05
47.87
88.22
26.95
RUS boosted treess
88.01
45.36
61.78
72.60
170
L. Singh et al.
as positive. High sensitivity indicates few false negatives which illustrates few cases of disease are missed. The specificity in diagnosis indicates the person having no disease as negative. Likewise, the highest 48.42% specificity is achieved by Quadratic SVM. The percentage of specificity achieved over ILPD dataset is quiet low Linear SVM along with Medium and Coarse Gaussian SVM has achieved 100% precision and highest F1 Score of 83.28%. Figure 3 shows graphical representation of Table 4, a comparative performance of 9 classifiers over various measures. The performance values of different machine learning classifiers calculated over Accuracy and Training-time measures. Table 5 represents the accuracy achieved and the execution time taken by various classifiers for the prediction of liver disease.
Fig. 3 Classifier’s performance on various measures
Table 5 Classification results of classifiers on accuracy and training time
Model type
Classifiers
Accuracy %
SVM
Linear SVM
71.4
Quadratic SVM 70.8
Ensemble
Training time (sec) 1.76 9.59
Cubic SVM
69.0
51.1
Fine gaussian SVM
71.0
0.80
Medium gaussian SVM
71.4
0.63
Coarse gaussian 72.2 SVM
0.61
Boosted trees
66.2
7.49
Bagged trees
70.7
4.98
RUS boosted trees
66.7
5.32
Classification of Hepatic Disease Using Machine Learning Algorithms
171
RUS Boosted Trees Bagged Trees Boosted Trees Coarse Gaussian SVM Medium Gaussian SVM Fine Gaussian SVM Cubic SVM Quadratic SVM Linear SVM 0
10
20
30
Training Time (sec)
40
50
60
70
80
Accuracy %
Fig. 4 Classifier’s performance on accuracy and training time
Figure 4 shows graphical presentation of comparative performance of 9 classifiers over accuracy and training time. It is clearly visible from the Table 5 that Coarse Gaussian SVM achieves the highest accuracy of 72.2% among all classifiers with minimum training time of 0.61 s. Confusion Matrix and ROC (Receiver Operating Characteristic) Area of Coarse Gaussian SVM is depicted in Fig. 5. Parallel coordinate plot is used to visualize high dimensional data, where each observation is represented by the sequence of its coordinate values plotted against their coordinate indices. For Coarse Gaussian SVM it is depicted in Fig. 6. After performing the experiment, Table 5 concludes that Coarse Gaussian SVM outperforms among all classifiers using the ILPD dataset by achieving an accuracy of 72.2% using minimum training time of 0.61 s.
Fig. 5 ROC and confusion matrix of coarse gaussian SVM
172
L. Singh et al.
Fig. 6 Parallel coordinate plot of coarse gaussian SVM
5 Conclusion The main strength of the designed model is the use of Machine learning algorithms used for classification in this study which were never been used to predict Liver disease. Though Liver diseases are hard to predict so this research work focuses on the key role of classification algorithms in predicting liver diseases. Classification algorithms are being chosen as they gain their strength in classifying the instances correctly once the system is trained while requiring less human efforts. Two classification models with their corresponding classifiers are used in conducting the research. For experimentation purpose Indian Liver Patient Database is chosen which is taken from UCI machine learning repository. Performances of all the classifiers are measured on various factors to fulfill the objective of research. Obtained results conclude Coarse Gaussian SVM as the best classifier as it achieves optimum accuracy among all the employed classifiers with minimum training time. Hence, Coarse Gaussian SVM outperforms in Liver Disease Diagnostic. Some more classifiers need to be explored for further research with the same objective.
Classification of Hepatic Disease Using Machine Learning Algorithms
173
References Alpaydin E (2010) Introduction to Machine Learning Second Edition Bahramirad S, Mustapha A, Eshraghi M (2013) Classification of liver disease diagnosis : a comparative study. IEEE, pp 42–46 Bendi Venkata Ramana, M. Surendra Prasad Babu NBV (2018) ILPD (Indian Liver Patient Dataset) Data Set. 1–2 Zhang C, Ma Y (2012) Ensemble machine learning. Springer International Publishing, New York Dordrecht Heidelberg London Dinu AJ, Ganesan R, Joseph F, Balaji V (2017) A study on deep machine learning algorithms for diagnosis of diseases. 12:6338–6346 Ghosh SR, Waheed S (2017) Analysis of classification algorithms for liver disease diagnosis. J Sci Technol Environ Inf 5:361–370 Janghel RR, Shukla A, Verma K (2016) Soft computing based expert system for Hepatitis and liver disorders. In: 2016 IEEE International Conference on Engineering and Technology (ICETECH). pp 740–744 Kidd PM (1996) Phosphatidylcholine, a superior protectant against liver damage. Altern Med Rev 1:258–274 Kumar A, Sahu N (2017) Categorization of Liver Disease Using Classification. 5:826–828 Maldini G (2014) What are the Signs of Liver Disease. [Online]. Available: https://www.hawaiipac ifichealth.org/media/3429/liver-disease-flyerr1.pdf Rajeswari P GSR (2010) Analysis of Liver Disorder Using Data mining algorithm. Glob J Comput Sci Technol 10:48–52 Patro S, Sahu KK (2015). Normalization: a preprocessing stage. arXiv preprint arXiv:1503.06462 Priya MB, Juliet PL, Tamilselvi PR (2018) Performance analysis of liver disease prediction using machine learning algorithms. Int Res J Eng Technol 5:206–211 Rahman A, Tasnim S (2014) Ensemble classifiers and their applications. A Review Abstract 10:31– 35 Sisodia D, Shrivastava SK, Jain RC (2010) ISVM for face recognition. In: Proceedings—2010 International Conference on Computational Intelligence and Communication Networks, CICN 2010. IEEE, pp 554–559 Sisodia D, Singh L, Sisodia S (2014) Fast and accurate face recognition using SVM and DCT. IEEE, pp 1027–1038 Sontakke S, Jay Lohokare RD (2017) Diagnosis of liver diseases using machine learning. International Conference on Emerging Trends & Innovation in ICT (ICEI). IEEE, Pune, pp 129–133 Vijayarani S, Dhayanand S (2015) Liver disease prediction using SVM and Naïve Bayes algorithms. Int J Sci Eng Technol Res 4:816–820
Anti-hyperlipidemic and Antioxidant Activities of a Combination of Terminalia Arjuna and Commiphora Mukul on Experimental Animals Jhakeshwar Prasad, Ashish Kumar Netam, Trilochan Satapathy, S. Prakash Rao, and Parag Jain Abstract The present study has been undertaken to evaluate antihyperlipidemic and antioxidant activities of Terminalia Arjuna (TA) and Commiphoramukul (CM) standardized extracts in combination at their predetermined doses. The hyperlipidemia in animals (rats) was induced by high fat diet by mixing Indian vanaspati ghee and coconut oil in the ratio of 3:1 (v/v). Acute toxicity was performed according to Organization of Economic Cooperation and Development (OECD) -423 and observed for behavioral changes, hematological and biochemical alteration if any or not for 14 days. The result of toxicity studies did not indicate any major changes in the result in comparison with control group of animals. The combination of plant extracts exhibited significant antihyperlipidemic activity in comparison to control group. The level of triglycerides (TGL), cholesterol (CHO), low-density lipoprotein (LDL), and very-low-density lipoprotein (VLDL) got reduced with increased level of highdensity lipoprotein (HDL). The result of test drug and standard drug showed similar value with some minor difference. Hematology data of hyperlipidemic rats showed safety level of blood components after treatment with test drug. The test drug also revealed good antioxidant activity by normalization of superoxide dismutase (SOD) and nitric oxide (NO) levels. Thus, further study required to determine the active constituents from plants extracts required for biological activities. Keywords Terminalia arjuna · Commiphoramukul · Super oxide dismutase · Nitric oxide
1 Introduction Higher amount of lipids or fats in the blood is characterized as hyperlipidemia. It is a family disorder in which fatty contents get increased abnormally. However, increased amount of fats increases the risk of coronary heart disease (CHD) and also plays role in body’s metabolic processes. Individual’s diet also shows impact on J. Prasad (B) · A. K. Netam · T. Satapathy · S. Prakash Rao · P. Jain Department of Pharmacology, Columbia Institute of Pharmacy, Tekari, Raipur, CG, India e-mail: [email protected] © Springer Nature Singapore Pte Ltd. 2021 A. A. Rizvanov et al. (eds.), Advances in Biomedical Engineering and Technology, Lecture Notes in Bioengineering, https://doi.org/10.1007/978-981-15-6329-4_16
175
176
J. Prasad et al.
hyperlipidemia; high cholesterol diet and food containing more saturated fats lead to increased blood cholesterol, and triglycerides levels. Other disorders, such as diabetes mellitus, kidney disease, and hypothyroidism, may promote hypertriglyceridemia (Fronzo and Ferrannini 1991). Most people who have hyperlipidemia also having relevant other complications such as diabetes, high cholesterol and have difficulty in managing all three conditions at a time. Hence combination therapies of more than two or three drugs are prescribed by physicians or clinicians those in turn produce severe adverse effects. Calcium channel blockers (CCBs) are one of the most potentially lethal prescriptions, which may worsen hyperlipidemia if administered excessively (Saeed and Larik 2017). Frequent administration of CCBs may cause rapid fall in blood pressure, decreased heart rate, and cardiac arrest. However, overdoses of sustained-release formulations result in delayed onset of dysrhythmias, shock, sudden cardiac collapse, and bowel ischemia. Among the anti diabetic agents, Di-Peptidyl Peptidase-IV (DPP-IV) inhibitors are drug of choice for treatment of Type-II diabetes and recent research revealed that, long term administration of DPP-IV inhibitors at their therapeutic doses also causes pancreatic cancers. Herbal drug had been used since ancient times for welfare of the mankind and several research have been done to identify the active compound responsible for therapeutic activity. The active components of plants when taken together may give synergistic effect,when they have co-administered for the treatment of multifactorial disorders such as diabetes associated with hypertension and dyslipidemia. Terminalia Arjuna(TA) is a wild herb containing various chemical constituents. Among these arjunetin and arjunosides acts as a major constituent already been reported for having affinity for Na+ - K+ ATPase Pump. (Urizar and Moore 2003) Commiphoramukul (CM) also reported for having antihyperlipidemic activity. The objective of this present research work is to evaluate the affinity of Terminalia arjuna for Na+ - K+ ATPase Pump blocking effect which in turn may be useful as an antihypertensive agent. When more than two drugs are administered at a time there may be a chance of drug interactions. Hence, toxicity studies need to be carried out for these combination therapies and to achieve better therapeutic response of Terminalia arjuna and Commiphora mukul standardized extract at their predetermined ratio for their synergistic effect in the treatment of high fat diet induced hyperlipidemia using experimental animals in Rats (Dobrian et al. (2000).
2 Materials and Methods 2.1 Drug and Chemical Reagents Terminalia Arjuna and Commiphoramukuldried extracts were received as a gift sample from SUNPURE Pvt. Ltd New Delhi (India). Carboxy methyl cellulose (0.5–5%) was purchased from LOBA Chemie Pvt. Ltd. Mumbai. Atorvastatin was procured from Sun Pharmaceuticals Pvt. Ltd. Mumbai, Maharashtra,
Anti-hyperlipidemic and Antioxidant Activities …
177
(India). Halothane was purchased from Korten Pharmaceuical Pvt. Ltd. ShantiSthal, Shirgaon-Palghar, Thane-Mumbai (India), and Formaldehyde was purchased from Merck Life Science Pvt. Ltd., Vikroli East, Mumbai, Maharashtra. All other chemicals used was of highest analytical grade-commercially available. Experimental Animals. Healthy adult Male Albino Wistar rats weighing 180–200 gm were obtained from the Animal House Facility of Columbia Institute of Pharmacy, Raipur, Chhattisgarh, (India) having certificate number CIP/IAEC/2017/103 and Regd. No.1321/PO/ReBi/S/10/CPCSEA. The animals were kept and maintained under controlled environmental conditions with temperature (23 ± 2 °C), relative humidity (40–50%), and 12/12 h light/dark cycle. The animals received a standard pellet diet (Hindustan lever limited, India) and water ad libitum. The animals used in the present study were cared as per the principles and guidelines of Institutional Animal Ethics Committee (IAEC), and in accordance with the CPCSEA, New Delhi, India. The animals were acclimatized to laboratory conditions for at least seven days before initiation of the experiment. Acute Toxicity. The acute toxicity was evaluated as per OECD guideline-423. Animals were received dose of Terminalia arjuna along with Commiphora mukul 250 mg/kg body weight orally administered by using an oral feeding needle after short fasting period. The general behavior of the animals was continuously monitored for 30 min, 1, 2, and 3 h after dosing, periodically during the first 24 h (with special attention given during the first 4 h) and then daily observed for 14 days. Experimental Study. The experiment was carried out on animals (albino Wistar rats) to determine therapeutic effectiveness of combination study. In this experiment rats of either gender were randomly divided into four groups. Each group consists of five animals either gender (n = 5). All the animals were administered high fat diet for induction of hyperlipidemia (Table 1). Induction of hyperlipidemia in rats. Hyperlipidemia was induced by feeding rats on diet rich in fats. It was prepared by mixing India Vanaspati ghee and coconut oil (3:1, v/v). This diet was given per-oral to rats at a dose of 3 ml/kg body weight daily (Munshi et al. 2014). Table 1 Allocation of animals into various groups for therapeutic effectiveness study
Groups
Treatment
Doses
1
Control group
Drinking water (Oral)
2
Toxic group
High fat diet (3 ml/kg)
3
Standard drug (Atorvastatin)
10 mg/kg
4
Test group [CM+ TA (50:50)]
500/kg/body weight (Oral)
178
J. Prasad et al.
Hematological Study. The blood was collected with EDTA anticoagulant through retro orbital puncture for biochemical estimation. The evaluated blood parameters were red blood cell count (RBC), blood hemoglobin concentration, basophil, eosinophil and neutrophil granulocytes, lymphocytes, and monocytes, hematocrit value, mean corpuscular volume (MCV), mean corpuscular hemoglobin (MCH), mean corpuscular hemoglobin concentration (MCHC), white blood cell (WBC), and platelet counts.
2.2 Biochemical Estimation of Antioxidants Superoxide Dismutase (SOD) Assay. The assay was performed by the production of superoxide from oxygen molecule using reduced b-nicotinamide adenine dinucleotide (NADH) as a reductant and phenazine methosulphate (PMS) as a catalyst. Nitrobluetetrazolium (NBT) was used as an indicator that turned blue when reduced by superoxide. Change in color was monitored spectrophotometrically in the visible range at 560 nm. While adding test drug to the reaction; the antioxidants (superoxide scavengers) competed with NBT to react with superoxide. The percent inhibition of NBT reduction was used to quantify superoxide-scavenging. Procedure. 10% w/v tissue homogenate in 0.15 M TrisHCl or, 0.1 M phosphate buffer was prepared and centrifuged at 15,000 rpm for 15 min at 4 °C. The supernatant (0.1 ml) was taken and considered it as sample. Then 0.1 ml sample + 1.2 ml sodium pyrophosphate buffer (pH 8.3, 0.052 M) + 0.1 ml phenazinemethosulphate (186 µM) + 0.3 ml of 300 µM Nitroblutetrazolium + 0.2 ml NADH (750 µM) were mixed and incubated at 30 °C for 90 s followed by addition of 0.1 ml glacial acetic acid. This was then stirred with 4.0 ml n-butanol and allowed to stand for 10 min followed by centrifugation, and butanol layer was separate. The Optical Density (OD) of the rest of the sample was measured at 560 nm by taking butanol as blank (Paoletti et al. 1986; Sapakal et al. 2008). Nitric Oxide Estimation. Nitric oxide is produced due to oxidative stress occurring in the brain. The assay was performed by taking 100 µl of serum sample in a test tube and added 400 µl of carbonate buffer (pH 9.0) followed by addition of copper cadmium alloy fillings (0.15 g). The reaction was stopped by addition of sodium hydroxide (100 µl of 0.35 M) and zinc sulphate solution (400 µl of 120 mM) under vortex mixing. Then the solution was allowed to stand for 10 min and centrifuged at 4000 rpm for 10 min. The clear supernatant solution (500 µl) was transferred to another test tube in which 500 µl of Griess reagent was added. The absorbance was noted spectrophotometrically at 548 nm. A standard curve (1–100 µM) was plotted using sodium nitrite to calculate the concentration of nitrite (Giustarini et al. 2008; Sastry et al. 2002). Procedure. Mix the following in a spectrophotometer cuvette (1 cm pathlength) i.e., 100 µL of Griess reagent, 300 µL of the nitrite-containing sample and 2.6 mL of deionized water. Then incubated the mixture for 30 min at room temperature. A
Anti-hyperlipidemic and Antioxidant Activities …
179
photometric reference sample was prepared by mixing 100 µL of Griess reagent and 2.9 ml of deionized water. Measured the absorbance of the nitrite-containing sample at 548 nm relative to the reference sample. Absorbance readings were converted to nitrite concentrations as described in calibration. Histopathological Examination. The animals were anaesthetized with halothane and blood was collected by retro orbital puncture for biochemical estimation. The animals were again anaesthetized by using excess halothane and sacrificed by cervical dislocation method. The abdominal portions were cut opened and heart was dissected out. The Heart was removed immediately and transferred into 10% formalin solution for routine histopathological examination. The samples were taken from the sections of rat heart tissue with highest macroscopic damage. The heart tissue specimen from each animal was removed and fixed in 10% formalin solution then cut into 5 µm thickness, stained using hematoxylin eosin for the histopathological examination. They were made using a rotary microtome, 5 µm thickness sections were cut from the tissue samples embedded in paraffin and placed on standard glass slides. The paraffin was melted with a period of approx 12 h in an incubator at 58 °C. The samples were then stained with haematoxylene and eosin (H&E) according to the protocol. Qualitative analyses were performed on 400× magnified images.
3 Results 3.1 The Effect of Terminalia Arjuna Along with Commiphora Mukul on Behavioral Changes The results of oral acute toxicity study indicated minor behavioral changes and no mortality observed in animals through the 3-days period following single oral administration at all selected dose levels of the Terminalia arjuna along with Commiphora mukul (Table 2 and Fig. 1).
3.2 The Effect of Terminalia Arjuna Along with Commiphora Mukul on Hematological Changes See Table 3.
180
J. Prasad et al.
Table 2 Effect of Terminalia arjuna along with Commiphora mukul on lipid profile level in Albino Wistar rats Parameters
Duration
Sex
Control
Toxic group
Standard
Test
TGL (mg/dL)
0 Day
M
62.5 ± 0.26
78.4 ± 0.22
63 ± 0.34
60.6 ± 0.33
F
63.4 ± 0.36
77.3 ± 0.20
64 ± 0.32
59.4 ± 0.37
10th Day
M
75.5 ± 0.23
89 ± 0.21
87.9 ± 0.24
73 ± 0.25
F
76.6 ± 0.34
88 ± 0.19
88.7 ± 0.27
72 ± 0.28
0 Day
M
53 ± 0.23
74 ± 0.25
57.2 ± 0.34
51.5 ± 0.36
F
54 ± 0.25
73 ± 0.24
58.5 ± 0.32
50.4 ± 0.34
M
56.2 ± 0.34
71 ± 0.31
70.1 ± 0.33
54.5 ± 0.23
F
57.2 ± 0.37
70 ± 0.30
71.3 ± 0.35
53.3 ± 0.25
0 Day
M
11.4 ± 0.25
3 ± 0.27
11.5 ± 0.22
12.3 ± 0.33
F
12.3 ± 0.34
2 ± 0.25
12.2 ± 0.20
12.7 ± 0.2
10th Day
M
13 ± 0.23
4 ± 0.21
18.3 ± 0.31
17.5 ± 0.27
F
14 ± 0.34
3 ± 0.20
19.4 ± 0.33
18.4 ± 0.25
M
23.6 ± 0.32
37 ± 0.30
22.5 ± 0.24
23.8 ± 0.21
F
24.5 ± 0.31
36 ± 0.28
23.3 ± 0.25
24.3 ± 0.23
10th Day
M
28.9 ± 0.25
26.7 ± 0.23
34.8 ± 0.32
44.8 ± 0.34
F
29.7 ± 0.27
25.5 ± 0.21
35.5 ± 0.33
45.7 ± 0.32
0 Day
M
15.2 ± 0.31
28 ± 0.29
15.7 ± 0.25
14.4 ± 0.31
F
14.5 ± 0.33
27 ± 0.27
16.5 ± 0.23
13.2 ± 0.33
M
15.8 ± 0.21
13.7 ± 0.19
17.9 ± 0.33
18.5 ± 0.34
F
16.6 ± 0.24
12.6 ± 0.17
18.7 ± 0.31
17.7 ± 0.32
CHO (mg/dL)
10th Day HDL (mg/dL)
LDL (mg/dL)
VLDL(mg/dL)
0 Day
10th Day
Mean ± SEM (n = 5) Triglycerides (TGL), Cholesterol (CHO), High density lipoprotein (HDL) Low density lipoprotein(LDL), Very-low-density lipoprotein (VLDL)
3.3 Biochemical Parameters Studies Superoxide Dismutase Assay See Table 4 and Fig. 2. Nitric Oxide (NO) Assay See Table 5 and Figs. 3 and 4. Histopathological Examination of Heart. Animal organ (Heart) histopathology report is shown below (Plates 1, 2, 3, 4, 5 and 6).
Anti-hyperlipidemic and Antioxidant Activities …
181
Animal groups
Lipid profile (mg/dl) 100 90 80 70 60 50 40 30 20 10 0
Control Toxic group Standard
M F M F M F M F M F M F M F M F M F M F 0 Day
10thDay 0 Day 10thDay 0 Day 10th Day 0 Day 10th Day 0 Day 10th Day
Time duration Fig. 1 Graph showing the effect of various treatments on lipid profile in different group of animals. All values are reported as Mean ± SEM (n = 5)
4 Discussion Hyperlipidemia is a multifactorial disorder involving interactions among environmental, vascular, neuroendocrine, and genetic factors. The prevalence of hyperlipidemia is increasing in India as well as all over the world. Apart from these, the other cause include is more complex i.e., association of type-2 diabetes mellitus as well as obesity. Those are polygenic factors. This complexity makes it difficult to diagnose the disorder properly that make the researchers to look major contributions toward the developments of new drug/new entity for effective treatment. As the drugs available in the market for the treatment of hyperlipidemia associated with diabetes are limited, many patients need the combination therapy of anti-lipidemics that in turn causes various side effects. Hence the herbal therapy has come into existence. Terminalia arjuna has traditionally been used for the treatment of various heart disorders for more than centuries. It improves cardiac muscle function subsequently improving pumping activity of the heart. Among the active constituent present in the Terminalia arjuna the saponin glycoside thought to be responsible for the ionotropic effect while flavonoids and oligomeric proanthocyanidins (OPCs) provide free radicals antioxidant activity. In other way Commiphora mukul an oleo gum-resin has been used as medications since Vedic period for the effective treatment of number of vascular disorders such as atherosclerosis, hypercholesterolemia, obesity, etc., but the scientific evidence for the combination of these two (Terminalia arjuna and Commiphora mukul) has not been established till yet. So this present study has been undertaken to evaluate the safety and effectiveness of both the drug at their
182
J. Prasad et al.
Table 3 Effect of combination therapy on hematological data of various groups of animal S. no. Particulars
Sex Control Group
Toxic Group
Standard group Test Group
1
Heamoglobin (gm%)
M
16.22 ± 0.08
5.21 ± 0.04
15.3 ± 0.07*
F
15.26 ± 0.107
4.20 ± 0.03
14.46 ± 0.120* 10.4 ± 0.141
Total WBC Count (cmm)
M
4280 ± 37.41
1120 ± 31.21 4030 ± 50.99
F
4100 ± 70.710 1090 ± 29.19 3880 ± 135.64
5010 ± 86.023*
Neutrophils (%)
M
61.4 ± 0.50
24 ± 0.39
58.8 ± 0.8
57.2 ± 0.8
F
59.4 ± 0.748
23 ± 0.37
57.4 ± 0.927
56.2 ± 0.8
Lymphocytes (%)
M
33.8 ± 0.37
15 ± 0.35
31.6 ± 0.50
F
32 ± 0.707
14 ± 0.32
5
Eosinophils (%)
M
6.2 ± 0.37
2 ± 0.31
F
4.6 ± 0.509
1 ± 0.30
6
Monocytes (%)
M
03 ± 00
0.3 ± 0.25
F
02 ± 00
0.2 ± 023
7
Basophiles (%)
M
00 ± 00
00 ± 00
00 ± 00
00 ± 00
F
00 ± 00
00 ± 00
00 ± 00
00 ± 00
8
RBC Count (%)
M
8.302 ± 0.00
1.3 ± 0.27
6.766 ± 0.00
5.694 ± 0.03
F
7.286 ± 0.012
1.1 ± 0.24
Platelet Count (%)
M
3.728 ± 0.00
0.2 ± 0.00
2.354 ± 0.00
2.174 ± 0.01
F
2.726 ± 0.009
0.1 ± 0.01
1.33 ± 0.010
1.34 ± 0.018
Mean Platelet M Value F (Million/cmm)
10.28 ± 0.09
1.65 ± 0.07
8.722 ± 0.15
8.502 ± 0.14
9.38 ± 0.106
1.35 ± 0.05
7.56 ± 0.107
7.46 ± 0.145
Packed Cell M Volume F (Million/cmm)
41.74 ± 0.05
21 ± 0.03
39.24 ± 0.09*
37.36 ± 0.10
40.42 ± 0.106
20 ± 0.01
Mean Corpuscular Volume (Cu micron)
M
50.548 ± 0.00
14 ± 0.04
48.57 ± 0.00
57.584 ± 0.00
F
47.76 ± 1.788
13 ± 0.02
47.5 ± 0.141
56.55 ± 0.014
Mean Corpuscular Hemoglobin (Pictograms)
M
19.28 ± 0.06
2.65 ± 0.07
17.602 ± 0.00* 16.742 ± 0.00*
F
18.42 ± 0.106
1.58 ± 0.05
16.54 ± 0.012
15.74 ± 0.014
Mean Corpuscular Hemoglobin Con. (mg/dl)
M
38.174 ± 0.00
14 ± 0.09
36.4 ± 0.50
35.29 ± 0.00*
F
37.18 ± 0.012
13 ± 0.07
35.2 ± 0.860
34.75 ± 0.018
Red Cell Distribution Width (%)
M
15.62 ± 0.05
5 ± 0.04
13.64 ± 0.05
11.36 ± 0.10*
F
14.36 ± 0.107
4 ± 0.02
12.5 ± 0.141*
10.5 ± 0.141
2 3 4
9 10
11
12
13
14
15
Mean ± SEM (n = 5), P = < 0.005 (*)
30.6 ± 0.927 4.4 ± 0.50 3 ± 0.707 02 ± 0.70 01 ± 0.583
5.73 ± 0.010
11.28 ± 0.10* 5060 ± 143*
30 ± 0.70* 28.2 ± 0.860 3.4 ± 0.87 3.1 ± 0.860 02. ± 0.45 0.1 ± 0.43
4.75 ± 0.014
38.5 ± 0.141* 36.6 ± 0.114
Anti-hyperlipidemic and Antioxidant Activities …
ConcentraƟon of SOD level (in nM/mL)
Table 4 Superoxide dismutase levels in different groups of animal
45 40 35 30 25 20 15 10 5 0
183
S. no.
Groups
SOD level
1
Control
42.00 ± 0.690
2
Toxic group
11.25 ± 1.24
3
Standard
39.41 ± 0.597
4
Test group
27.47 ± 1.194
Control
Toxic group
Standard
Test group
Treatment groups Fig. 2 Graph showing the SOD levels in homogenized heart tissue of different groups of animal Table 5 Nitric oxide levels in different groups of animal
S. no.
Groups
NO level
1
Control
13.42 ± 0.144
2
Toxic group
3
Standard
11.67 ± 0.382
4
Test group
10.17 ± 0.289
2.12 ± 0.683
Absorbance
Standard curve of NO 0.6 0.5 0.4 0.3 0.2 0.1 0
y = 0.0047x + 0.0444 R² = 0.9983 0
20
40
60
80
Concentration Fig. 3 Graphical representation of standard curve of NO (Nitric Oxide)
100
120
184
J. Prasad et al. 16 14 12 10 8 6 4 2 0 Control
Toxic group
Standard
Test group
Fig. 4 Graph showing the levels of NO homogenized heart tissue of different groups of animal
Plate no. 1 Sample preparation of heart control group
predetermined dose level ratio by using high cholesterol diet hyperlipidemia in rat model. The oral acute toxicity study for combination of both drugs in rats was carried out. The results for the acute toxicity study indicated that, there were no morbidity and mortality in animals of all the groups. The combination of drugs exhibited decreased level of TGL, CHO, LDL, and VLDL but increased level of HDL. Thus, representing
Anti-hyperlipidemic and Antioxidant Activities … Plate no. 2 Effect of vehicle on histopathological changes of heart tissue in control group
Plate no. 3 Sample preparation of heart test group
185
186 Plate no. 4 Effect of Terminalia arjuna along with Commiphora mukul histopathology report of heart test group
Plate no. 5 Sample preparation of heart standard Group
J. Prasad et al.
Anti-hyperlipidemic and Antioxidant Activities …
187
Plate no. 6 Effect of Atorvastatin histopathology report of heart standard group
antihyperlipidemic effect in comparison to control group. The results of antioxidant activity (SOD and NO level) revealed that the combination therapy showed good antioxidant activity on 10th day. Further, exhaustive study is required to determine active constituents and establish the exact mechanism responsible for biological activities.
5 Conclusions In this present study, various parameters were evaluated for establishment of safety and effectiveness of combination therapy containing Terminalia arjuna and Commiphora mukul. Both the drugs in combination with their predetermined ratios exhibited significant antihyperlipidemic and antioxidant properties. The result of oral acute toxicity study did not show any behavioral changes and mortality.
References De Fronzo RA, Ferrannini E (1991) Insulin resistance: a multifaceted syndrome responsible for NIDDM, obesity, hypertension, dyslipidemia, and atherosclerotic cardiovascular disease. Diabetes Care 1;14(3):173–194; Horowitz BZ (2001) Toxicity, calcium channel blocker. E-Medicine: Instant Access to the Minds of Medicine, September 6, 2001 Dobrian AD, Davies MJ, Prewitt RL, Lauterio TJ (2000) Development of hypertension in a rat model of diet-induced obesity. Hypertension 35(4):1009–1015
188
J. Prasad et al.
Giustarini D, Rossi R, Milzani A, Dalle Donne (2008) IBT-M in E. Nitrite and nitrate measurement by Griess reagent in human plasma: evaluation of interferences and standardization. In: Nitric Oxide, Part F [Internet]. Academic Press, p 361 Munshi RP, Joshi SG, Rane BN (2014) Development of an experimental diet model in rats to study hyperlipidemia and insulin resistance, markers for coronary heart disease. Indian J Pharmacol 46(3):270 Paoletti F, Aldinucci D, Mocali A, and Caparrini A, (1986) A sensitive spectrophotometric method for the determination of superoxide dismutase activity in tissue extracts. Anal Biochem 153:536– 541 Saeed A, Larik FA (2017) Current Topics in herbal medicine with applications in cardiovascular diseases. Cardiovasc Dis 1(1):257 Sapakal VD, Shikalgar TS, Ghadge RV, Adnaik RS, Naikwade NS, and Magdum CS (2008) In vivo screening of antioxidant profile: a review. J Herb Med Toxicol 2(2):1–8 Sastry KV, Moudgal RP, Mohan J, Tyagi JS, Rao GS (2002) Spectrophotometric determination of serum nitrite and nitrate by copper-cadmium alloy. Anal Biochem 306:79–82 Urizar NL, Moore DD (2003) GUGULIPID: a natural cholesterol-lowering agent. Annu Rev Nutr 23(1):303–313
Epileptic Seizure Detection Using Deep Recurrent Neural Networks in EEG Signals Archana Verma and Rekh Ram Janghel
Abstract Epilepsy is a neurological ailment that influence around 1% of mankind. Around 10% of the United States population experience at least a single convulsion in their life. Epilepsy is distinguished by the inclination of the brain to generate unexpected bursts of strange electrical action which disrupts the normal functioning of the brain. Generally patients not really recognises their conditions so mostly experts suggests that electroencephalography (EEG) for seizure detection. In this research, we implement Recurrent Neural Networks (RNNs) based on Gated Recurrent Unit (GRU) with and without Wavelet Filter technique for early seizure detection. In this paper work in 5-layer of GRU Recurrent Neural Networks (RNNs) technique is implemented to distinguish healthy and seizure class. The proposed system accomplished an accuracy of with DWT 98.50% and without 97.0%. Keywords Deep neural networks · Epilepsy · EEG · Electroencephalogram · GRU · RNN · Recurrent neural networks · Seizure detection
1 Introduction Epilepsy is a association of typical neurological conditions of a mind distinguished by recurrent unprovoked convulsions which are a consequence of sudden bursts of abnormal electrical discharges in the brain (Guo et al. 2012; Orosco et al. 2013; Fayek et al. 2017). According to report by WHO (Orosco et al. 2013) around 50 million people are influenced by epilepsy worldwide (Talathi 2017; Acharya et al. 2013; Cook 2013; Guo et al. 2010a). Nearly one in every 100 persons are influenced by a convulsion in their lifespan (Guo et al. 2011), approximately 2.4 million new cases of epilepsy are reported every year globally (Acharya et al. 2013; Geetha 2012). Then, A. Verma (B) · R. R. Janghel Department of Information Technology, National Institute of Technology, Raipur, India e-mail: [email protected] R. R. Janghel e-mail: [email protected] © Springer Nature Singapore Pte Ltd. 2021 A. A. Rizvanov et al. (eds.), Advances in Biomedical Engineering and Technology, Lecture Notes in Bioengineering, https://doi.org/10.1007/978-981-15-6329-4_17
189
190
A. Verma and R. R. Janghel
the manifestation of an epileptic is eccentric and its movement of activity has been less understood (Guo et al. 2011), approximately 2.4 million new cases of epilepsy are reported every year globally (Acharya et al. 2013; Iasemidis 2003). Epileptic patients experience varying symptoms depending on the section and the expanse of the brain that is affected, Epileptic seizures can induce gloomy physical, social consequences, and psychological, which include loss of consciousness, injury, and abrupt death. Epilepsy is of two types depending upon the degree of involvement of the brain tissue, which are, generalized seizures and partial seizures, generalized seizures roughly involve nearly the complete brain, partial seizures develop in a particular section of the brain and remain limited to that section (Guo et al. 2010a). Electroencephalography (EEG) is a famous electrophysiological technique to comprehend the complicated action of human brain (Gandhi et al. 2010). EEG directly gauges and registers the electrical action of the mind. Spontaneous EEG signals are classified into several rhythms based on their frequencies, which are band (3–4 Hz); band (4–8 Hz); band (8–13 Hz); band (13–30 Hz) (Zhou and J. 2004; Parvez and Paul 2016). An EEG is especially useful at times when the brain is at risk by providing a sensitive indication of cerebral functioning. It is long time ranges signals consequently. An extended EEG recording is required early examinations for demonstrated proof of this anomalous movement to be an advantage to detect in epilepsy and cerebral tumors. These days EEG signals are utilized to get data relevant to the diagnosis, prognosis, and treatment of these abnormal conditions. EEG is registered using electrodes placed on the scalp and have tiny amplitudes of the range of 20 V (Selvan and Srinivasan 1999; Ammar and Senouci 2016). The electrodes are placed as per the 10–20 international system which has been shown in the Fig. 1. Usually EEGs contain massive amounts of information and detection of traces of epilepsy requires a visual inspection of the total span of the EEG by a specialist Fig. 1 Standardized electrode placement scheme (Talathi 2017)
Epileptic Seizure Detection Using Deep Recurrent Neural Networks …
191
which is a cumbersome task (Guo et al. 2010b; Lin et al. 2016). Hence, developing automated epileptic seizure espial system is noteworthy for reviewing EEGs. Over a past two decades, many algorithms have been applied for epileptic seizure detection, which include time–frequency analysis methods, non-linear statistical models, and more present day machine learning methodologies, for example, neural systems and Support Vector Machines (SVM), however inspite of many Progress, current EEG analysis approach are a long way from expert with majority of the methods being considered because of their high false detection rates. Moreover, Sabrina Ammar [34] have work on EEG signals seizer detection in single-channel with the help of extreme learning machine. Also NihalFatmaGler use EEG signals classification for employing Lyapunov exponents using RNN. Tawfik et al. (2016) dose a hybrid automated system use for epileptic seizures detection in help of EEG signals. Kumar et al. (2008) work with Recurrent Neural Network Classifier use for Epilepsy Seizure detection using an automated Wavelet Entropy for feature selection. A novel deep structure was currently presented which attempts to minimize the false alarm rate on EEG signals. This framework incorporates CNNs with RNNs to convey the state of the art performance and use for enhancing accuracy.
2 Methodology 2.1 Dataset A data elucidated by Andrzejak et al. (2001) was employed for the current research. The total data set is comprised of 5 subsets, (marked as Z, O, N, F, S) every subset having 100 single-channel EEG fragments each being on 23:6 s time span sampled at 173:6 Hz. The fragments were chosen from regular multi-channel EEG records after visual examination for artifacts for e.g., because of muscle action or eye movements. Z, O sets contain fragments that have been obtained from EEG recordings conducted on 5 healthy people. People were in an awaken condition and eye open (Z) and eye closed (O). Sets N, F, and S are emerged from an EEG. Fragmentation set F has been registered for the epileptogenic zone, and fragment in set is hippocampal. To do so, set N and F encompassed only activity measured during seizure free spans, set S contains only seizure activity.
2.2 Discrete Wavelet Transform Analyzing Wavelet transform is a spectral estimation procedure in which any general function can be communicated as an endless arrangement of wavelet. Behind of this idea analysis contain communicating a signal as a linear combination of a specific arrangement of set obtained by shifting and dilating one single function
192
A. Verma and R. R. Janghel
called a mother wavelet. The disintegration of the signal prompts an arrangement of coefficient called wavelet coefficient. So that a signal can be reformed a linear combination of the wavelet functions weight by as a wavelet coefficient. The key element of wavelets is the time–frequency limitation (Subasi 2007b). It implies that a large portion of the vitality of the wavelet is confined to a limited time interim. Frequency limitation implies that the Fourier change is band constrained. EEG signal will uncover highlights identified with the transient nature of the signal which is not clear by the Fourier transform. The separated wavelet coefficients give a minimal illustrates that demonstrates the vitality dissemination of the EEG signal in time and frequency (Cho et al. 2014).
2.3 Recurrent Neural Networks (RNNs) With refined recurrent hidden units like the Gated Recurrent Unit (GRU) and the Long-Short-Term-Memory (LSTM) have turned into a mainstream decision for modeling temporal sequences (Chung et al. 2014; Che et al. 2018), Conventional Feed Forward Neural Networks are trained on labeled data till the prediction error is minimized (Gandhi et al. 2010), whereas the decision of RNN at time t depends on the decision at which the RNN reached at time step t-1. Inspired by the success of LSTMs Chung et al. (2014), proposed a network architecture which is simpler and efficient compared to LSTMs known as the Gated Recurrent Unit (GRU) (Che et al. 2018) the architecture of a GRU has been depicted, Example of a RNN is shown in Fig. 2. We used the Keras library for implementation of Gated Recurrent Unit (GRU), 60% of the information was utilized for preparing and 40% of information is utilized for approval. Lately, Recurrent Neural Networks (RNNs) with sophisticated recurrent hidden units like the GRU and the LSTM have turned to a famous choice for modeling temporal sequences (Talathi 2017; Graves 2013; Fayek et al. 2017; Sathyanarayana et al. 2016), Conventional Feed Forward Neural Networks are trained on labeled data till the prediction error is minimized (Fig. 3).
Fig. 2 Recurrent neural networks (Gandhi et al. 2010)
Epileptic Seizure Detection Using Deep Recurrent Neural Networks …
193
Fig. 3 Raw EEG data belonging to each of subsets a Z, b O, and c S
The decision of RNN at time t depends on the decision at which the RNN reached at time step t-1. In particular we focus on the 2-class classification of a given EEG fragment into either healthy or, ictal states. As subset consists of 100 EEG segments, from each of the subjects 60% of the data is used to prepare training and the remainder is used for validation. Hence the training and testing data consists of 150 segments.
2.4 Gated Recurrent Unit (GRU) GRU was put forward by Graves et al. (2013) so as to enable every recurrent unit adaptively capture the dependencies among various time range. GRU, similar to LSTM consists of gating units which regulate the flow of data within unit, however j it doesn’t consist of independent memory cells. The activation h t of the GRU at time j t directs initiation between the before activation h t−1 and the candidate activation j h˜ t−1 : h t = (1 − z t )h (t−1) + z t h˜ t j
j
j
j
j
(1)
j
The update gate z t determines the amount by which the unit updates its activation. j
z t = σ (W z xt + U z h (t−1) ) j
(2)
194
A. Verma and R. R. Janghel
The candidate activation h˜ t is computed similarly to the update gate: j
j h˜ t = tanh(Wxt + U(rt ∗ h (t−1) )) j
(3)
j
The reset gate rt is computed in a similar way: j
rt = σ (W r xt + U r h (t−1) ) j
(4)
Algorithm 1: Pseudocode for Gated Recurrent Unit (GRU) Input: Data dictionaries each embodying 100-single 4097 samples Output: accuracy, precision, recall, F-Score Step.1. Read Z001 - Z100; Read O001 - O100; Read S001 - S100; Step.2. Create model using Keras library comprising of five layers as listed in Table. 1 Step.3. Split the data in the ratio 60:40 for training and testing Step.4. Train the model using 60% data using model.fit() function Step.5. Test the model using 40% data using model.predict() function. Step.6. Determine the precision, recall, F-Score using the sklearn library. Step.7. Determine the confusion-matrix with the help of mlxtend Step.8. Calculate accuracy from the confusion-matrix. end In Fig. 4 we firstly take raw EEG signals set of Z, S, and O and we do shuffling then split data as 60% for training and 40% for testing then perform RNN Model (GRU) then done classification as normal and seizure classes. Secondly we use DWT as feature extraction and get better performance and work for RNN Model (Table 1). Prediction: the following parameters were used to evaluate our proposed method. Precision (P) =
TP (T P + F P)
TP (T P + F N ) 1 + β 2 (P)(R) F-Score = 2 β ·P·R Recall (R) =
TP = True Positive FN = False Negative FP = False Positive
(5) (6)
(7)
Epileptic Seizure Detection Using Deep Recurrent Neural Networks …
195
Raw EEG
DWT
Without DWT
Split data
Training
Testing
RNN Model (GRU)
Classification
Normal
Seizure
Fig. 4 Flowchart of proposed methodology
Table 1 Parameter Design for GRU
TN = True Negative P = Precision R = Recall.
Layer type
Output shape
Parameters
Input
(51, 80, 1)
0 30,600
GRU 0
(51,100, 1)
Fc
(51, 100, 1)
10,100
GRU 1
(51, 100)
60,300
LR
(51, 3)
303
196 Table 2 Performance measurement using deep RNN with DWT
A. Verma and R. R. Janghel DWT Levels
Epochs
Leaning rate
Accuracy (%)
D1
1–1000 Epochs
0.01– 0.09
91.20
D2
1–1000 Epochs
0.01–0.09
95.40
D3
1–1000 Epochs
0.01–0.09
97.00
D4
1–1000 Epochs
0.01–0.09
96.10
D5
1–1000 Epochs
0.01–0.09
98.50
3 Result and Discussion 3.1 Result We check all D1, D2, D3, D4, and D5 featured data set for varying (0–0.09) learning rate and number of epochs to test our proposed method and best result includes in Table 2.
4 Comparison with Other Work Multiple methods have been put forward for epileptic seizure detection. Table demonstrates a differentiation between various methods, methods assessed on the same data set have been incorporated so that a distinction between the results is realistic.
5 Conclusion and Future Work EEG signs can be utilized to separate among ordinary and epileptic conditions of the cerebrum. In this work, we put forward an epileptic seizure detection system built on Deep Recurrent Neural Networks using Gated Recurrent Unit (GRU) which can be trained with high accuracy and classify the EEG segments as either healthy or ictal. Our purposed method is capable of achieving an accuracy of up to 97% and with DWT comes 98.50%. The performance of purposed model is better than few reckon in that Table 3. The benefit of this proposed model presented in this paper is large data set. In this proposed method may require more diversity of data to optimum performance. In this paper, we can improve performance of our purposed method by increasing the number of samples. The performance of this method can be improved by increasing the number of samples.
Epileptic Seizure Detection Using Deep Recurrent Neural Networks …
197
Table 3 Synopsis of erstwhile research for detection of epileptic and normal classes Authors
Classifiers
Accuracy (%)
Kannathal et al. (2005)
Diagnostic neural Network
97.20
Sadati et al. (2006)
Adaptive Neuro-Fuzzy Inference systems (ANFIS)
92.20
Subasi (2007a)
Adaptive neural fuzzy network
85.90
Guo et al. 2009)
Mixture expert model (a modular neural network)
94.50
Cho et al. (2014)
Feed Forward ANN
95.20
Proposed method
Deep Recurrent Neural Networks (RNN) based on Gated Recurrent Unit (GRU)without DWT
97.00
Proposed method
Deep Recurrent Neural Networks (RNN) based on Gated Recurrent Unit (GRU) with DWT
98.50
References Acharya UR, Sree SV, Swapna G, Martis RJ, Suri JS (2013) Automated EEG analysis of epilepsy: a review. Knowl-Based Syst 45:147–165 Ammar S, Senouci M (2016) Seizure detection with single-channel EEG using extreme learning machine. In: 2016 17th international conference on sciences and techniques of automatic control and computer engineering (STA), pp 776–779 Andrzejak RG, Lehnertz K, Mormann F, Rieke C, David P, Elger CE (2001) Indications of nonlinear deterministic and finite-dimensional structures in time series of brain electrical activity: dependence on recording region and brain state. Phys Rev E 64(6):61907 Che Z, Purushotham S, Cho K, Sontag D, Liu Y (2018) Recurrent neural networks for multivariate time series with missing values. Sci Rep 8(1): 6085 Cho K, Van Merriënboer B, Bahdanau D, Bengio Y (2014) On the properties of neural machine translation: Encoder-decoder approaches. arXiv1409.1259 Chung J, Gulcehre C, Cho K, Bengio Y (2014) Empirical evaluation of gated recurrent neural networks on sequence modeling. arXiv1412.3555 Cook MJ et al (2013) Prediction of seizure likelihood with a long-term, implanted seizure advisory system in patients with drug-resistant epilepsy: a first-in-man study. Lancet Neurol 12(6):563–571 Fayek HM, Lech M, Cavedon L (2017) Evaluating deep learning architectures for speech emotion recognition. Neural Netw 92:60–68 Gandhi T, Panigrahi BK, Bhatia M, Anand S (2010) Expert model for detection of epileptic activity in EEG signature. Expert Syst Appl 37(4):3513–3520 Geetha G (2012) Detecting epileptic seizures using Electroencephalogram: a novel technique for seizure classification using fast walsh-hadamard transform and hybrid extreme learning machine. In: Proceedings of the second international conference on computational science, engineering and information technology, pp 253–260 Graves A (2013) Generating sequences with recurrent neural networks. arXiv1308.0850 Graves A, Mohamed A, Hinton G (2013) Speech recognition with deep recurrent neural networks. In: 2013 IEEE international conference on acoustics, speech and signal processing (ICASSP), pp 6645–6649 Guo L, Rivero D, Seoane JA, Pazos A (2009) Classification of EEG signals using relative wavelet energy and artificial neural networks. In: Proceedings of the first ACM/SIGEVO summit on genetic and evolutionary computation, pp 177–184 Guo L, Rivero D, Dorado J, Rabunal JR, Pazos A (2010a) Automatic epileptic seizure detection in EEGs based on line length feature and artificial neural networks. J Neurosci Methods 191(1):101– 109
198
A. Verma and R. R. Janghel
Guo L, Rivero D, Pazos A (2010b) Epileptic seizure detection using multiwavelet transform based approximate entropy and artificial neural networks. J Neurosci Methods 193(1):156–163 Guo L, Rivero D, Dorado J, Munteanu CR, Pazos A (2011) Automatic feature extraction using genetic programming: an application to epileptic EEG classification. Expert Syst Appl 38(8):10425–10436 Guo P, Wang J, Gao XZ, Tanskanen JMA (2012) Epileptic EEG signal classification with marching pursuit based on harmony search method. In 2012 IEEE international conference on systems, man, and cybernetics (SMC), pp 283–288 Iasemidis LD et al (2003) Adaptive epileptic seizure prediction system. IEEE Trans Biomed Eng 50(5):616–627 Kannathal N, Lim M, Acharya UR, Sadasivan PK (2005) Entropies for detection of epilepsy in EEG Kumar SP, Sriraam N, Benakop PG (2008) Automated detection of epileptic seizures using wavelet entropy feature with recurrent neural network classifier. In: TENCON 2008–2008 IEEE region 10 conference, pp 1–5 Lin Q et al. (2016) A novel approach for epileptic EEG signals classification based on biclustering technique. In: 2016 international conference on machine learning and cybernetics (ICMLC), vol 2, pp 756–760 Orosco L, Correa AG, Laciar E (2013) A survey of performance and techniques for automatic epilepsy detection. J Med Biol Eng 33(6):526–537 Parvez MZ, Paul M (2016) Epileptic seizure prediction by exploiting spatiotemporal relationship of EEG signals using phase correlation. IEEE Trans Neural Syst Rehabil Eng 24(1):158–168 Sadati N, Mohseni HR, Maghsoudi A (2006) Epileptic seizure detection using neural fuzzy networks. In: 2006 IEEE international conference on fuzzy systems, pp 596–600 Sathyanarayana A et al (2016) Impact of physical activity on sleep: a deep learning based exploration. arXiv1607.07034 Selvan S, Srinivasan R (1999) Removal of ocular artifacts from EEG using an efficient neural network based adaptive filtering technique, vol 6, no 12, pp 330–332 Subasi A (2007a) EEG signal classification using wavelet feature extraction and a mixture of expert model. Expert Syst Appl 32(4):1084–1093 Subasi A (2007b) EEG signal classification using wavelet feature extraction and a mixture of expert model. Expert Syst Appl 32(4):1084–1093 Talathi SS (2017) Deep Recurrent Neural Networks for seizure detection and early seizure detection systems. arXiv1706.03283 Tawfik NS, Youssef SM, Kholief M (2016) A hybrid automated detection of epileptic seizures in EEG records. Comput Electr Eng 53:177–190 Zhou W, Gotman J (2004) Removal of EMG and ECG artifacts from EEG based on wavelet transform and ICA. In: 26th Annual International Conference of the IEEE Engineering in Medicine and Biology Society, 2004, vol 1, pp 392–395. IEMBS’04
Detection of Disease from Leaf of Vegetables and Fruits Using Deep Learning Technique Avisha Jaiswal, Saurabh Pathak, Yogesh Kumar Rathore, and Rekh Ram Janghel
Abstract In a country like India where 16% of the total GDP growth is contributed by agriculture alone and majority of the people rely on it for their source of living, it is of utmost importance that the threats to the production of crops should be minimized. Crop leaf disease is major type of diseases suffered by crops; its manual identification is a difficult task leading to wrong treatment and poor production. It gives rise to the need of an accurate automated system to detect plant leaf diseases, and, is made possible by the recent advances in computer vision and deep learning. In this paper we use Convolution Neural Networks to classify and identify vegetable leaf diseases. An open database of Plant Village that contains 54, 306 plant images, with 26 classes of diseases and 14 different crop species is used to curate a new database in reference to our problem domain which contains 5 class of diseases suffered by vegetables having images of diseased and healthy leaves. We have implemented CNN models, namely, Sequential and GoogLeNet with images of leaves as input. The highest success rates achieved are 98.48% and 97.47% for Sequential and GoogLeNet model, respectively. The relatively higher accuracy of the models makes them very useful and eligible to be used to solve the current problem of crop disease in India. It could further be used to make a better vegetable disease identification system. Keywords GoogleNet · Sequential model · Leaf disease · Plant village dataset · CNN
A. Jaiswal (B) · S. Pathak · Y. K. Rathore · R. R. Janghel National Institute of Technology, Raipur, CG, India e-mail: [email protected] S. Pathak e-mail: [email protected] Y. K. Rathore e-mail: [email protected] R. R. Janghel e-mail: [email protected] © Springer Nature Singapore Pte Ltd. 2021 A. A. Rizvanov et al. (eds.), Advances in Biomedical Engineering and Technology, Lecture Notes in Bioengineering, https://doi.org/10.1007/978-981-15-6329-4_18
199
200
A. Jaiswal et al.
1 Introduction India is an agricultural country with 70% of its population relying on agriculture hence, it becomes vital, in a country like India, that cultivation of crops gives maximum yield. Crop health is affected by environmental degradation, pests, chemicals and pesticides and various other factors making it a major challenge in agricultural sector, and the losses to the country’s economy and farmers due to it are miserable. Identification of crop disease by optical observation of symptoms is a tedious and complex task leading to mistaken conclusions (Konstantinos and Ferentinos 2018). Faster and accurate predictions of leaf diseases in plants could help to develop an early treatment technique, while considering reducing economic losses (Akila et al. 2018) which could be achieved by the existence of an automated system adding valuable assistance to the task of diagnosis (Konstantinos and Ferentinos 2018)). The simplicity and ease of the system would lead it to be used by the farmers as a valuable tool in the parts of the county with lesser awareness and lacking infrastructure. The recent developments advances in the domain of computer vision backed-up by deep learning has paved way for machine-controlled disease identification (Mohanty and Marcel 2016). Deep learning constitutes a recent modern technique for image processing and data analysis with accurate results (Suk and Shen 2013). The applicability of deep learning is versatile and it has been applied to almost every problem domain including agriculture (Konstantinos and Ferentinos 2018). We have used deep learning for identification and classification of plant leaf disease.
1.1 Deep Learning Deep learning is a modern branch of machine learning and a subset of artificial intelligence which is inspired from human brain (Suk and Shen 2013) and represents better representational power for future representation. It can be thought as of novel approach in learning representations from data that gives importance on learning of consecutive layers (Chollet 2018). Deep learning neural networks contain many hidden layers which enable them to effectively extract high level features (Suk and Shen 2013). An error of 16.4% was reached by a deep CNN architecture for the classification of pictures into 1,000 probable classes, which was top 5 in 2012 (Krizhevsky et al. (2012). In the following three years, the error rate has been lowered to 3.57% by a variety of architectures in deep convolutional neural networks (Krizhevsky et al. 2012; Zeiler and Fergus 2014; Simonyan and Zisserman 2014). The feasibility of state-of-art deep learning models, which have made computation much easier, have left revolutionary impact on sectors like recognition and processing of (Cunn et al. 1681), voice recognition (Srivastava and Hinton 2014), and alternative equally advanced issues that require working with enormous amount of data, giving a big boom to use of such technologies in areas automation of vehicle driving, artificial
Detection of Disease from Leaf of Vegetables and Fruits …
201
intelligence, and inference, etc. The introduction of these deep learning methodologies in the field of agriculture (Carranza-Rojas et al. 2017), and particularly within the field of disease identification, has taken place only since last two extents and the extent is very limited our problem domain specializes in classifying and identifying the disease in leaves of vegetables specifically. Five diseases, namely, Bacterial Spot, Black Rot, Late Blight, Early Blight, and Powdery Mildew, from which various vegetable leaves, suffer are selected and classes for these diseases are made. A database is created having healthy and diseased images of each class of disease, the images were selected from Plant Village dataset that contains large number of images containing different disease (Hughes and Salathé 2015). An accuracy of 98.48% is achieved for sequential model and is maximum so far, GoogLeNet gave an accuracy of 97.47%.
2 Literature Review The very basic deep learning methodology that has been used throughout this work is Convolution Neural Network (CNN) (Konstantinos and Ferentinos 2018). CNN is a deep learning architecture which is inspired by the visual perception of living beings (Gu et al. 2018). The problems which are complex and are acquainted with large volume of data and pattern recognition, CNN serve as a powerful methodology (Konstantinos and Ferentinos 2018). For the task of identification and classification of plant leaf disease we have used Sequential and GoogLeNet architectures for CNN. Grinblat et al. (2016) proposed a model which can easily detect three different spices of plant using a powerful neural network (Grinblat et al. 2016). A relatively smaller work was carried out by Sladojevic et al. using same source of data but only for 13 diseases and 5 plants (Sladojevic et al. 2016). The accuracy of these two was in a range of 91–98% depending upon training testing set split. More recently, comparisons were made between some image processing based techniques and deep learning-based techniques which uses different dataset of plant leaves and fruits by Pawara et al. (2017), and the results of CNN beating the performance of those conventional methods were drastic. Finally, models for detection of 9 different tomato diseases and pests were developed by Fuentes et al. (2017) whose performance was satisfactory (Fuentes et al. 2017). “Konstantinos P. Ferentinos compared the performance of various CNN architectures out of which GoogLeNet gave an accuracy of 97.27% high (Konstantinos 2018). Mohanty et al. compared AlexNet and GoogLeNet architectures of CNNs for detection of plant diseases, dataset of leaves images containing 26 plant diseases for 14 different crops was used “in which GoogLeNet achieved a maximum accuracy of 96.21% for a 80 isto 20 train test split in contrast to our work in which GoogLeNet has achieved a success rate of 97.47% for same test train split” (Mohanty et al. 1419). “Sequential architecture so far, has been used for other purposes achieving all time high accuracy of 95.61% for filling of slots in languages which are spoken (Ngoc 2018) whereas our sequential model has achieved a success rate of 98.48%.”
202
A. Jaiswal et al.
3 Methodology 3.1 Data Acquisition To address the problem of classification and identification of vegetable leaf disease Plant Village (Mohanty et al. 2016) database for images containing 54, 306 images of plant leaves with 26 classes of diseases and 14 different crop species is used to create a new dataset. We have selected 5 classes of diseases for all the vegetable species in Plant Village dataset. The details regarding the classes of diseases and respective number of images in each class are as follows: Disease Bacterial Spot: 5123 images Black Rot: 1801 images Early Blight: 1966 images Late Blight: 2909 images Powdery Mildew: 2887 images Healthy: 8680 images. Since the images obtained from this source were already preprocessed, hence all the preprocessing tasks, labeling, and augmentation are ignored to curate the database. To avoid the issue of over fitting complete database was divided into training dataset and testing dataset in a ratio of 80 is to 20. Other such similar divisions like 70 is to 30 should not have any effects on final classification results.
3.2 Methodology Artificial Intelligence. The study of imitating and automating human thinking is artificial intelligence and the mathematical models via which it can be achieved are artificial neural networks (Konstantinos and Ferentinos 2018). Artificial neural networks have synapse that interconnects their neurons via which they mimic the functioning of brain (Konstantinos and Ferentinos 2018). The main characteristics of artificial neural networks are their ability to learn by themselves through supervised learning and training. They are trained on some dataset that contains certain specific matchings of the input and output data regarding the system which we want our artificial neural network to model. Convolutional Neural Networks. Convolutional Neural Networks are evolutionary and transformed version of artificial neural networks (LeCun and Bengio 1995), where main feature is that they have a number of layers and reduce the requirement for number of artificial neurons to a very large extent in comparison to normal artificial neural networks (Konstantinos and Ferentinos 2018) and “applies” the same for image recognition purpose. CNN specializes in tasks related to recognition of images hence various models of CNN have been developed for image recognition.
Detection of Disease from Leaf of Vegetables and Fruits …
203
The two architectures that we have used in our work are Sequential and GoogLeNet model. Sequential Architecture. Sequential model is the most basic type of model for Convolutional Neural Network. The layers in this model are simply stacked and sequentially arranged (see Fig. 1) (Ngoc 2018). The layers have input, output, input_shape, output_shape. Each of the layers of sequential model has a defining configuration. In our implementation of sequential model there are a total of 8 layers in which 4 are convolutional and 4 are max-pooling and these are alternatively arranged. To build a sequential model add layers to it one by one and compile the model with a loss function, an optimizer, and optional evaluation metrics. Then use the dataset to fit the model (Ngoc 2018). GoogLeNet Architecture. GoogLeNet architecture is a very deep and wide architecture with 22 layers. The specialty of GoogLeNet is that even after having a wide and deep architecture the number of parameters used in it is considerably low (Hughes and Salathé 2015). The peculiar feature of GoogLeNet architecture is that it’s a type of network architecture (see Fig. 2) (LeCun et al. 2015) which are in the form of inception module. 1X1, 3X3, and 5X5 convolutional and a max-pooling layer are used in parallel by inception module which helps it to capture a range of characteristics in parallel. Output of all these parallel layers is concatenated by a filter concatenation. GoogLeNet is made by 9 such inception modules (Going et al. 2015).
Fig. 1 Architecture of sequential model (Ngoc 2018)
Fig. 2 GoogLeNetArchitecture (LeCun et al. 2015)
204
A. Jaiswal et al.
4 Result Performance of both the models is assessed by training the model on the dataset curated from Plant Village database.The parameters that have been varied in order to assess the performance of architectures are batch size and number of epochs, as can be seen in Table 1. In GoogLeNet for batch size 32 success rate kept on increasing as we increased the number of epochs and touched an all time high when number of epochs were 30, then decreased exponentially. Similar phenomenon occurred for Sequential model at when the batch size was 32 and numbers of epochs were 30. So the maximum success rates achieved are 98.48% and 97.47% in the case of sequential and GoogLeNet architectures, respectively, for batch size 32 and number of epochs 30 in both cases (see Table 1). Here, x-axis shows the number of epochs (i.e. 10, 20, 30, 40) varying with two different batch size of input data i.e., 32 and 64 and y-axis shows the accuracy of particular model (see Fig. 3). Table 1 Table showing success rates by varying parameters Neural network
Architecture
Epoch
Batch sisze
Accuracy (Diseased vs. healthy) (%)
Convolution neural network
GoogLeNet
10
32
93.15
20
94.26
25
96.84
30 10
97.47 64
20
94.12
25
95.73
30 Convolution neural network
Sequential
92.26
10
91.39 32
93.62
20
91.65
25
96.36
30
98.48
10
64
93.45
20
89.74
25
95.45
30
95.87
Detection of Disease from Leaf of Vegetables and Fruits …
205
Fig. 3 Result comparison
5 Conclusion In this work we used convolutional neural networks to develop and assess systems for classification and identification of vegetable leaf diseases “accurately, so as to solve the problem of misleading treatments which arise due to wrong identification of disease that further leads to huge losses in crop production.” The relatively higher accuracy of the models makes them very useful and eligible to be used to solve the current problem of crop disease in India. It could further be used to make a better vegetable disease identification system. The maximum success rates achieved are 98.48% and 97.47% in the case of sequential and GoogLeNet architectures, respectively.
References AkilaM, Deepan P (2018) Detection and classification of plant leaf diseases by using deep learning algorithm. In: International journal of engineering research & technology, ICONNECT conference proceedings, vol 6, issue 7 Carranza-Rojas J, Joly AAJ, Bonnet P, Hervé HG (2017) Goëau, Erick Mata-Montero. Automated Herbarium Specimen Identification using Deep Learning, In: Proceedings of TDWG Chollet F (2018) Title of deep learning with python Fuentes A, Yoon S, Kim SC, Park DS (2017) A robust deep-learning-based detector for real-time tomato plant diseases and pest recognition. Sensors 17:2022 Grinblat GL, Uzal LC, Larese MG, Granitto PM (2016) Deep learning for plant identification using vein morphological patterns. Comput Electron Agric 127:418–424 Gu J, Wang Z, Kuen J (2018) Recent advances in convolutional neural networks, pattern recognition, vol 77. Elsevier, pp 354–377 Hughes DP, Salathé M (2015) An open access repository of images on plant health to enable the development of mobile disease diagnosis Konstantinos P (2018) Ferentinos Deep learning models for plant disease detection and diagnosis. Comput Electron Agricul 145(2018):311–318
206
A. Jaiswal et al.
Konstantinos T, Ferentinos P (2018) Deep learning models for plant disease detection and diagnosis. Comput Electron Agricul 145:311–318 Krizhevsky A, Sutskever I, Hinton GE (2012) Imagenet classification with deep convolutional neural networks. In: Advances in neural information processing systems, pp 1097–1105 LeCun Y, Bengio Y (1995) Convolutional networks for images, speech, and time series. In: The handbook of brain theory and neural networks, vol 3361(10) LeCunn Y, Haffner P, Forsyth DA et al (1999) Object recognition with gradient based learning: shape, contour. LNCS 1681. Springer, Berlin, Heidelberg, pp 319–345 LeCun Y, Bengio Y, Hinton G (2015) Deep learning. Nature 521:436–444. https://doi.org/10.1038/ nature14539 Mohanty SP, Hughes D, Salathé M (2016) Using deep learning for image-based plant disease detection, pp 233–240 Mohanty SP, Hughes DP, Salathé M (2016) Using deep learning for image-based plant disease detection. Front Plant Sci 1419 Pawara P, Okafor E, Surinta O, Schomaker L, Wiering M (2017) Comparing local descriptors and bags of visual words to deep convolutional neural networks for plant recognition. In: 6th international conference on pattern recognition applications and methods (ICPRAM 2017) Simonyan K, Zisserman A (2014) Very deep convolutional networks for large-scale image recognition, pp 1409–1556 Sladojevic S, Arsenovic M, Anderla A, Culibrk D, Stefanovic D (2016) Deep neural networks based recognition of plant diseases by leaf image classification. In: Computational intelligence and neuroscience Srivastava N, Hinton N (2014) Dropout: a simple way to prevent neural networks from overfitting. J Mach Learn Res 15:1929–1958 Suk H-I, Shen D (2013) Deep learning-based feature representation for AD/MCI classification. Med Image Comput Comput Assist Interv 16(2):583–590 Szegedy C et al (2015) Going deeper with convolutions. In: Proceedings of the IEEE conference on computer vision and pattern recognition, pp 1–9 Vu NT (2018) Sequential standing. Deep learning with Python (2018) Zeiler MD, Fergus R (2014) Visualizing and understanding convolutional networks. In: Computer vision–ECCV, pp 818–833
Evaluation for Toxicity and Improved Therapeutic Effectiveness of Natural Polymer Co-administered Along with Venocin in Acetic Acid-Induced Colitis Using Rat Model Ashish Kumar Netam, Jhakeshwar Prasad, Trilochan Satapathy, and Parag Jain Abstract In these present study, two different series of experiments such as evaluation for toxicity and therapeutic of effectiveness of Aesculus hippocastanum coadministered low viscosity of sodium alginate has been carried out by using rat model (Acetic acid-induced ulcerative colitis). Acute and subacute toxicities were performed according to OECD guideline-423 and 407, respectively. In vitro antioxidant study for the combination therapy was studied. Histopathological examinations were carried out to determined organ level toxicity on long-term use of such combination. The results for In vitro antioxidant study suggested the free radical scavenging activity of the combination therapy. There were no behavioral changes or any morbidity and mortality were observed during the oral acute toxicity study food intake, water intake, and body weight variation in all the group of animals were within the similar pattern that means no much changes has been observed. Among the animals between control tests and standard drug-treated groups. The result of histopathological data indicated that some changes were observed for hemoglobin content, red blood cell count, etc., in some test group which is negligible in comparison with control group. The microscopic feature of histopathological study for the different organs such as kidney, liver, heart, etc., indicated that some degeneration and necrosis were observed in all the test groups of animals. These alterations of histopathological changes may be due to the stress, infection, and administration of test compound in empty stomach. Further study is suggested for determination of appropriate dose and ratios (for combination) to reduce the long-term toxicity and to improve the therapeutic effectiveness for the benefit of the entire society. Keywords Aesculus hippocastanum · Low viscosity sodium alginate · Acetic acid antioxidant · Ulcerative colitis · Lipidperoxidase · Tumor Necrosis Factor (TNF-α)
A. K. Netam (B) · J. Prasad · T. Satapathy · P. Jain Department of Pharmacology, Columbia Institute of Pharmacy, Tekari, Raipur, CG, India e-mail: [email protected] © Springer Nature Singapore Pte Ltd. 2021 A. A. Rizvanov et al. (eds.), Advances in Biomedical Engineering and Technology, Lecture Notes in Bioengineering, https://doi.org/10.1007/978-981-15-6329-4_19
207
208
A. K. Netam et al.
1 Introduction Inflammatory bowel disease (IBD) is an intestinal disorder leads to inflamed and ulcerative intestine. It includes Crohn’s disease and ulcerative colitis. Ulcerative colitis affected rectal and colonic mucosa, (Carter et al. 2004.) This is including Crohn’s disease and ulcerative colitis which is an idiopathic inflammatory bowel disease of the rectal and colonic mucosa (Morris et al. 1989). It is characterized by colonic inflammation, resulting most probably from the infiltration of polymorphonuclear cells, lymphocytes, monocytes, and plasma cells, accompanied by the overproduction of oxygen free radicals, ultimately leading to mucosal alteration and ulceration (Almenier et al. 2012). Ulcerative colitis is a chronic disease cause disturbance in homeostasis in the gastrointestinal tract and intestinal inflammation (Baumgart and Carding 2007). It is also affecting the mucosal layer of the distal colon and rectum. The main symptoms of ulcerative colitis include diarrhea, abdominal cramps, and recurrent blood in the stools caused by mucosal ulcers. Ulcerative colitis about 50 lakh people have affected by inflammatory bowel disease across the world and India. Annually, 12 lakh cases of IBD have been reported in India but unfortunately only few people are aware about the disease (Lennard-Jones 1989). The recent pharmacological therapy for patients with ulcerative colitis includes nonselective anti-inflammatory drugs and corticosteroids or immunosuppressants, as well as anti-TNF-α agents (Cho et al. 2007). Several abovementioned drugs are used for the treatment of ulcerative colitis but they possess several adverse effects. Hence herbal remedies came into existence as an alternative therapy to overcome the disadvantages of such drug (Kane et al. 2003). These drugs are used to maintain continuing long-term remission, reduction of abnormal colonic inflammation, and control of clinical symptoms, such as diarrhea, rectal bleeding, and abdominal pain (Mowat et al. 2011). Though, the continuous use of these medications can cause serious side effects to patients. Thus, a great effort has been made to develop new drugs to treat ulcerative colitis. It has been reported that herbal drugs are the best alternative for the treatment of ulcerative colitis. Herbal drugs are safe in comparison to other existing drugs, these can be used to prevent long-term remission, reduction of colonic inflammation, controlling clinical symptoms including rectal bleeding, diarrhea, and abdominal pain. Therefore, to avoid serious adverse effects of allopathic drugs, greater efforts have been paying to develop new drugs to treat ulcerative colitis. The natural herbal drugs Aesculus hippocastanum is used by several researchers for the effective treatment of deep vein thrombosis and other venous disorders which is generally seen in ulcerative colitis by considering the therapeutic effectiveness of Aesculus hippocastanum. Hence we have decided to use Aesculus hippocastanum as a test substance to evaluate its potency. The Sodium alginate is well known as biocompatible, degradable, and nontoxic. It forms a gel without the need of heat. They are also widely used as protective reparative effects. So, in this research we have proposed to administer the natural polymer sodium alginate alone and in combination with Aesculus
Evaluation for Toxicity and Improved Therapeutic Effectiveness …
209
hippocastanum in their appropriate ratio to determine the toxicity and therapeutic effectiveness against Acetic acid-induced ulcerative colitis in rat model.
2 Materials and Methods 2.1 Drug and Chemical Reagents Aesculus hippocastanum was received as a gift sample from SUNPURE Pvt. Ltd New Delhi (India). Sodium Alginate was obtained from SD FINE Chem Ltd. Mumbai (India). Glacial acetic acid (AA) 99.8% was purchased from LOBA Chemie Pvt. Ltd. Mumbai (India). Sulfasalazine was procured from WALLACE Pharmaceuticals Pvt. Ltd. Ponda, Goa Maharashtra, (India). The RayBio® Enzyme-linked immunosorbent assay (ELISA) kits for rat TNF-α was obtained from Norcross USA. Lignocaine HCl Gel was purchased from ALVES Healthcare Pvt. Ltd. Mumbai, (India). All other chemicals used were of highest analytical grade commercially available.
2.2 Experimental Animals Healthy adult Male Albino Wistar rats weighing about 180–200 gm were obtained from the Animal House Facility of Columbia Institute of Pharmacy, Raipur, Chhattisgarh, (India) having certificate number CIP/IAEC/2017/102 and Regd. No. 1321/PO/ReBi/S/10/CPCSEA. The animals were kept maintained under controlled environmental conditions with temperature (23 ± 2 °C), relative humidity (40–50%), and 12/12 h light/dark cycle with unlimited access to standard pelleted diet (chow, food) and water ad libitum, as per CPCSEA guideline. The animals were acclimatized to laboratory conditions for at least seven days before initiation of the experiment.
2.3 Experimental Design The animals were randomly separated into five groups each containing six animals. Group I (Negative control) was pretreated with vehicle every 12 h, per oral; Group II (Toxic control) acetic acid, 2 ml; Group III (Test group-1) received AesculusHippocastanum 5 mg/kg oral; Group IV (Test group-2) received LVA 5gm/kg intrarectally; Group V (Test group-3) received 5 mg/kg mixture of Aesculus hippocastanum and LVA, intrarectally; Group VI (Reference group) was treated with Sulfasalazine 24 h before acetic acid instillation and for the subsequent five days.
210
A. K. Netam et al.
2.4 Acute Toxicity The acute toxicity was evaluated as per OECD guideline-423. The animals were randomly divided into five groups. They were received a dose of low viscosity sodium alginate along with Aesculus hippocastanum in Wistar rat of 50 mg/kg, body weight orally administered by using oral gavages after short fasting period. The general behavior of the animals was continuously monitored for 30 min, 1, 2, and 3 h after dosing, periodically during the first 24 h and the same treatment was followed for seven days.
2.5 Induction of Colitis Inductions of ulcerative colitis were used according to the method of GhasemiPirbaluti et al. (2017), with slight modification. The animals have fasted overnight with free access to water. The animals were light anesthetized with halothane. The inducing agent; acetic acid (2 ml, 1%, v/v) was instilled into the anus verge by inserting a medical grade polyurethane tube with 2 mm diameter through the rectum into the colon to a distance of 8 cm. The tube was kept in vertical position during instillation and after instillation to avoid leakage of acetic acid solution. Following the enema, After that, animals were kept in cages with continuous supply of feed and water till 8th day. Halothane was used to anesthetize animals and biochemical estimation was performed by collecting blood by retro-orbital puncture for biochemical estimation. The animals were again anaesthetized by using excess halothane and sacrificed by cervical dislocation. The abdominal portions were cut opened and colon was dissected out. Colon was flushed gently with saline and weighed. It was used for macroscopic scoring and histopathological estimations.
2.6 Hematological Study The blood was collected with EDTA anticoagulant through retro-orbital puncture for biochemical estimation. The evaluated blood parameters were red blood cell count, blood hemoglobin concentration, basophil, eosinophil and neutrophil granulocytes, lymphocytes, and monocytes, hematocrit, mean corpuscular volume (MCV), mean corpuscular hemoglobin (MCH), mean corpuscular hemoglobin concentration (MCHC), white blood cell, and platelet counts (Byelinska et al. 2018).
Evaluation for Toxicity and Improved Therapeutic Effectiveness …
211
2.7 Antioxidant Activity Lipid Peroxidase/Malonaldehyde (LPO/MDA) A colonic tissue sample was homogenized in potassium phosphate buffer (50 mMol/L, pH 7.4, 1 g tissue/5 mL buffer). The total amount of protein in each sample was measured using the Bradford method. The tissue homogenate (10% w/v) was prepared in 0.15 M Tris-HCl buffer (PH 7.4). Then to it 0.2 ml of 8.1% sodium dodecyl sulphate (SDS) + 1.5 ml 20% acetic acid + 1.5 ml 8% Thiobarbutric acid (TBA) were added and volume was made up to 4 ml with distilled water. The above solution was subjected to heat on water bath for 60 min using glass ball as condenser. Then the solution allowed cooling and volume was made up to 5 ml. Then 5 ml of butanol: pyridine (15:1) was added and vortexed for a period of 2 min followed by centrifuge at 3000 rpm for 10 min. The upper organic layer was taken and measured optical density was measured at 532 nm. The absorbance was considered as total malondialdehyde (MDA) formed (Bose et al. 1989; Hagar et al. 2007; Alam et al. 2013).
2.8 Measurement of TNF-α Colon was removed and homogenized in PBS then the amount of protein in each sample was measured via the Bradford method. The results were expressed in pg of cytokine/mg of protein. Assessment of cytokines (TNF- α) was carried out using ELISA kit; in clonic tissue, strips were minced with scissors for 15 s, suspended in 2 ml of 10 mm PBS (7.4 pH) and incubated in a shaking water bath 37 °C for 20 min. The sample was centrifuged and the supernatants were kept at −70 °C. The TNF-α assay using ELISA kit was performed (Wallace et al. 1989; Bose et al. 1989).
2.9 Histopathological Evaluation The samples of highest macroscopic damage were selected from the sections of rat colon tissue. A two cm portion of the colonic tissue specimen from each animal was removed and fixed in 10% formalin solution then cut into 5 μm thickness, stained using hematoxylin eosin for the histopathological examination. They were made using a rotary microtome, 5 μm thickness sections were cut from the tissue samples embedded in paraffin and placed on standard glass slides. The paraffin was melted with a period of approx 12 h in an incubator at 58 °C. The samples were then stained with haematoxylene and eosin (H&E) according to the protocol. Qualitative analyses were performed on 400× magnified images.
212
A. K. Netam et al.
Table 1 Observation table of animals behavioral Test
Gender
Control
Control Reversal
Test-1
Test-2
Test-3
Test-3 reversal
Tremor
M
–
–
–
–
–
–
F
–
–
–
–
–
–
Convulsion
M
–
–
–
–
–
–
F
–
–
–
–
–
–
M
–
–
–
–
–
–
F
–
–
–
–
–
–
Diarrhea
M
–
–
–
–
–
–
F
–
–
–
–
–
–
Sleep
M
–
–
–
–
–
–
F
–
–
–
–
–
–
Salivation
3 Result 3.1 Behavioral Changes The results of oral acute toxicity study indicated that behavioral changes were no mortality and morbidity observed in animals through the 3-days period following single oral administration at all selected dose levels of the low viscosity sodium alginate along with Aesculus hippocastanum (Table 1).
3.2 Body Weight Loss The result of body weight in different groups of animals reveled that there were no much changes have been observed (Table 2 and Fig. 1).
3.3 Hematological Study See Tables 3 and 4.
3.4 MDA Activity See Table 5 and Fig. 2.
F
183.6 ± 2.16 185.5 ± 2.96
185.2 ± 2.94
F 185.9 ± 2.96
185.4 ± 2.94
185.3 ± 2.92
M
183.4 ± 2.15
183.5 ± 3.05
183.2 ± 3.03
F
185.8 ± 2.98
184.8 ± 3.16
184.5 ± 3.15
M
M
182.3 ± 2.13
182.1 ± 2.12
F
Mean ± SEM (n = 6)
28 Days
21 Days
14 Days
181.5 ± 1.33 183.8 ± 2.15
181.2 ± 1.43
183.6 ± 2.13
F
M
182.6 ± 2.57
182.4 ± 2.97
M
0 Days
7 Days
Control reversal
Control
Gender
Time period
185.4 ± 2.29
185.5 ± 2.96
183.1 ± 2.07
185.1 ± 2.93
183.2 ± 3.04
184.2 ± 3.12
182.5 ± 2.11
183.5 ± 2.11
181.6 ± 1.54
182.4 ± 2.97
Test-1
185.2 ± 2.94
185.4 ± 2.94
183.8 ± 2.16
185.2 ± 2.94
183.3 ± 3.06
184.6 ± 3.15
182.5 ± 2.13
183.6 ± 2.12
181.5 ± 1.33
182.5 ± 2.51
Test-2
185.7 ± 2.97
185.6 ± 2.97
183.6 ± 2.12
185.5 ± 2.96
183.5 ± 3.07
183.8 ± 3.18
182.6 ± 2.11
183.7 ± 2.15
181.4 ± 1.25
182.6 ± 2.57
Test-3
Table 2 Effect of Low viscosity sodium alginate along with Aesculus hippocastanum on body weight in Albino Wistar Rats Test-3 reversal
185.9 ± 2.99
185.8 ± 2.99
185.8 ± 2.99
185.7 ± 2.97
183.7 ± 3.09
183.9 ± 3.19
182.9 ± 3.1
183.8 ± 2.18
181.4 ± 1.32
182.8 ± 2.98
Evaluation for Toxicity and Improved Therapeutic Effectiveness … 213
214
A. K. Netam et al. 187.3
Animal Groups
185.5 Control 183.8
Control Reversal Test-1
182.
Test-2
180.3
Test-3 178.5
Test-3 Reversal
176.8 0 Days
7 Days
14 Days
21 Days
28 Days
Time Duration
Body Weight (gm) Fig. 1 Graphical representation of animals mean body weight during dosing, number of animals per group n = 10, each group (Five—Male and Five—Female). All value is reported as mean ± SEM (n = 6) Table 3 Hematological data of various groups of male animals S. no.
Parameters
Control group
Test-1 group
Test-2 group
Test-3 group
1.
Hb (gm%)
16.48 ± 0.298
16.46 ± 0.102
16.26 ± 0.102
16.2 ± 0.2
2.
WBC (cmm)
2560 ± 40
3420 ± 152.9
4360 ± 112.2
4480 ± 106.77
3.
Neu (%)
42.4 ± 0.244
40.4 ± 0.812
43.8 ± 0.969
44 ± 00
4.
Lym (%)
52 ± 0.004*
53.8 ± 0.969
47.8 ± 0.734
47.2 ± 0.244
5.
Eos (%)
4.2 ± 0.374
5 ± 0.316
4.6 ± 0.244
4.4 ± 0.509
6.
Mon (%)
01 ± 00
01 ± 00
01 ± 00
01 ± 00
7.
Bas (%)
00 ± 00
00 ± 00
00 ± 00
00 ± 00
8.
RBC (%)
7.61 ± 0.002*
7.53 ± 0.046
7.54 ± 0.011
7.53 ± 0.009
9.
Platelet (%)
2.75 ± 0.003*
2.91 ± 0.019
2.94 ± 0.007
2.95 ± 0.013
10.
MPV
9.62 ± 0.058
9.46 ± 0.097
9.4 ± 0.054
9.5 ± 0.004*
11.
PCV
33.38 ± 0.165
33.40 ± 0.329
33.42 ± 0.631
33.46 ± 0.082
12.
MCV
51 ± 0.196
52.69 ± 0.087
52.72 ± 0.12
52.75 ± 0.01
13.
MCHb (Pictograms)
19.72 ± 0.009
19.68 ± 0.026
19.92 ± 0.002*
19.68 ± 0.058
14.
MCHb (mg/dl)
36.47 ± 0.006
37.35 ± 0.011
37.63 ± 0.022
37.66 ± 0.009
15.
RCDW (%)
15.22 ± 0.037
15.28 ± 0.086
15.5 ± 0.094
15.6 ± 0.005*
Mean ± SEM (n = 5), P value = < 0.005(*)
Evaluation for Toxicity and Improved Therapeutic Effectiveness …
215
Table 4 Hematological data of various groups of female animals S. no.
Parameters
Control group
Test-1 group
Test-2 group
Test-3 group
1.
Hb (gm%)
15.48 ± 0.298
14.46 ± 0.102*
14.26 ± 0.102
14.2 ± 0.2
2.
WBC (cmm)
2350 ± 41
2352 ± 42.3
2355 ± 42.2
2355 ± 43.7
3.
Neu (%)
42.4 ± 0.24
40.4 ± 0.23
43.8 ± 0.29
44 ± 00
4.
Lym (%)
51 ± 0.70
51.4 ± 0.69
52.4 ± 0.73
52.6 ± 0.74
5.
Eos (%)
4.2 ± 0.34
4.5 ± 0.36
4.6 ± 0.34
4.6 ± 0.39
6.
Mon (%)
01 ± 00
01 ± 00
01 ± 00
01 ± 00
7.
Bas (%)
00 ± 00
00 ± 00
00 ± 00
00 ± 00
8.
RBC (%)
7.61 ± 0.002
7.63 ± 0.046
6. ± 0.011
7.60 ± 0.009*
9.
Platelet (%)
3.75 ± 0.003
3.91 ± 0.019*
2.74 ± 0.007
3.14 ± 0.013
10.
MPV
9.62 ± 0.058
9.46 ± 0.097
10.3 ± 0.054
10.6 ± 0.050*
11.
PCV
47.38 ± 0.165
33.84 ± 0.329
38.22 ± 0.631
42.46 ± 0.082
12.
MCV
55.41 ± 0.196
51.22 ± 0.087
52.724 ± 0.12
49.938 ± 0.01
13.
MCHb (Pictograms)
19.72 ± 0.009
20.68 ± 0.026
19.92 ± 0.02*
18.68 ± 0.058
14.
MCHb (mg/dl)
35.47 ± 0.006
40.35 ± 0.011
37.73 ± 0.022
37.66 ± 0.009
15.
RCDW (%)
15.72 ± 0.037
16.28 ± 0.086*
16.5 ± 0.094*
16.36 ± 0.05*
Table 5 Serum MDA levels in different groups of animals
Fig. 2 Graph showing the level of serum MDA in homogenized colon tissue of different group of animals
S. no.
Groups
MDA level
1
Control
8.22 ± 0.509
2
Test-1
5.23 ± 0.152
3
Test-2
4.57 ± 0.052
4
Standard
4.78 ± 0.072
MDA level 11.25
1
2
3
4
9. 6.75 4.5 2.25 0. 1
2
3
4
216
A. K. Netam et al.
Table 6 TNF-α level in different groups
S. No.
Groups
TNF-α level
1
Control
5.53 ± 0.672
2
Test-1
4.84 ± 0.252
3
Test-2
4.59 ± 0.239
4
Standard
5.13 ± 0.584
Determination of TNF-α
7. 5.6
1 2
4.2
3
2.8
4
1.4 0. 1
2
3
4
Fig. 3 Graph showing the level of TNF-α in homogenized colon tissue of different group of animals
3.5 TNF-α Activity See Table 6 and Fig. 3.
3.6 Histopathology See Fig. 4.
4 Discussion The modern pharmaceutical research is concerned with all aspects of identifying new chemical substances with new modes of action. In particular, the economics of treatment linked to drug dosage has led compound to new drug development technologies. As a result, treatments are now becoming more reasonable for wide sections of society, including the financially challenged. Few marketed products are available for the effective treatment of IBD such as Sulfasalazine, Mesalazine, Balsalazide, Antileukotriene like Montelukast, etc. Natural polymers are evaluated for their wound-healing effect in case of IBD. Steroidal drugs such as prednisone,
Evaluation for Toxicity and Improved Therapeutic Effectiveness …
217
Fig. 4 a Control group. b Test group-1. c Test group-2. d Standard group
etc., on long-term use cause severe side effect like Cushing syndrome. Many herbal products possess anti-inflammatory, antioxidant activity has been evaluated for their anti-IBD effect. Venocin from Horse chestnut seed extract has been shown to have Antioxidant, Anti-inflammatory, Wound healing, and Supports circulation property. Several abovementioned allopathic drugs are used for the treatment of ulcerative colitis but they possess several adverse effects. Hence herbal remedies came into existence as an alternative therapy to overcome the disadvantages of such drugs. Among the natural herbal drugs Aesculus hippocastanum is used by several researchers for the effective treatment of deep vein thrombosis and other venous disorder which is generally seen in ulcerative colitis by considering the therapeutic effectiveness of Aesculus hippocastanum. We have decided to use Aesculus hippocastanum as a test substance to evaluate its potency against acetic acid-induced ulcerative colitis using rat model. Further it has been evidenced from the literature that sodium alginate is a natural polymer used for various gastrointestinal disorders though sodium alginate is obtained from natural source and devoid of any side effects/adverse effects. The Sodium alginate is well known as biocompatible, degradable, and nontoxic. It forms a gel without the need of heat. They are also widely used as protective reparative effects. Though we have taken the combination of sodium alginate which is co-administered with Venocin, there is a need to evaluate the toxicity and safety of the combination. Two series of studies have been carried out for the evaluation of toxicity and therapeutics effectiveness for the combination of Aesculus hippocastanum along with low viscosity sodium alginate at their predetermined dose.
218
A. K. Netam et al.
The oral toxicity study has been carried out as per OECD-423. The results of the oral acute toxicity study indicated that there were no mortality and morbidity observed in animal of all the groups. The result of body weight in different groups of animals reveled that there were no much changes has been observed similarly, food and water consumption all the group of animals showing similar pattern of result during the entire course of experiment the effect of combination therapy on Hematological data of all the groups of animal when depicted in Table no. 15 and Fig no. 6–18. The hemoglobin content except control and control reversal group of animals reduced to a lesser extent whereas in case of animal of test-3 reversal group the value has been increased up to 17.3. The WBC count in all groups of animals has been increased. To some extent in comparison to control and control reversal group neutrophils lymphocytes and eosinophils value did not indicated much change in all groups of animals. RBC count has been decreased in Test-1, Test-2 as well as Test-3 reversal group whereas Test-3 group of animals having similar RBC count value in comparison to control reversal group platelet count and mean platelet value shown no much change in all group of animals. The value for packed cell volume has been reduced in Test-1, Test-2, and Test-3 reversal group of animals. The results for Mean corpuscular volume having similar pattern of results in all groups of animals. The results of mean corpuscular Hemoglobin content indicated that it has been increased in all the test groups of animal in comparison to control and control reversal group. The antioxidant activity for the combination therapy has been determined by the estimation of MDA and TNF-α. The results for MDA have been depicted in table no 17, 18, and graphical represented in graph no. 20 and 21. The results indicated that Test compound showing decreased in concentration of MDA and TNF-α. Hence the combination therapy possesses free radical scavenging activity. On the termination of experiment the animal was sacrificed as per CPCSEA guidelines and Subjected organ where isolated and subjected for histopathological examination to determine the toxicity of combination therapy at different organ level. The histopathological indicated that on long-term use of combination therapy some degeneration and necrosis have been observed in Test group of animals whereas no changes in microscopic features were pointed out in control and control reversal group of animals. These may be due to the excessive stress less food intake, etc. From the above finding it has been observed that the combination of Aesculus hippocastanum and Low viscosity sodium alginate at predetermined dosed possess very good free radical scavenging activity and anti-inflammatory activity and TNF-α inhibiting activity. To reduce the long-term use of organ level toxicity, further study is suggested to adjust the dose level and duration which produce better therapeutic effectiveness which in turn pave the way for the development of new drug.
5 Conclusion The present study has been undertaken to establish the improved therapeutic effectiveness of Aesculus hippocastanum co-administered with LVA. From the previously published scientific data, Aesculus hippocastanum is used for the treatment of various
Evaluation for Toxicity and Improved Therapeutic Effectiveness …
219
venous disorders such as deep vein thrombosis, etc., and Sodium alginate is a natural polymer having good mucosal protective activity. Hence, the present study has been proposed to determine the toxicity and synergistic effect of both the drugs with their appropriate ratios. The study has been carried out using Wistar rats and acetic acid was used as inducing agent for ulcerative colitis. Various In vitro studies such as Malondialdehyde (MDA) carried out and the result revealed that the combination possess good level of antioxidant activity. Hematological and biochemical findings also support our hypothesis but long-term administration of combinations (AesculusHippocastanum along with low viscosity sodium alginate) alters the mucosal integrity which has been observed from histopathological findings of some vital organs. So, our findings suggest that further detailed study is required to establish the exact mechanism of mucosal degeneration which in turn reduces the long-term organ level toxicity that will help the researchers to decide for further new drug development for the benefit of the society.
References Alam MN, Bristi NJ, Rafiquzzaman M (2013) Review on in vivo and in vitro methods evaluation of antioxidant activity. Saudi Pharm J 21(2):143–152 Almenier HA, Al Menshawy HH, Maher MM, Al Gamal S (2012) Oxidative stress and inflammatory bowel disease. Front Biosci (Elite Ed) 1(4):1335–1344 Baumgart DC, Carding SR (2007) Inflammatory bowel disease: cause and immunobiology. Lancet 369(9573):1627–1640 Bose R, Sutherland GR, Pinsky C (1989) Biological and methodological implications of prostaglandin involvement in mouse brain lipid peroxidation measurements. Neurochem Res 14(3):217–220 Byelinska IV, Kuznietsova HM, Dziubenko NV, Lynchak OV, Rybalchenko TV, Prylutskyy YI, Kyzyma OA, Ivankov O, Rybalchenko VK, Ritter U (2018) Effect of C60 fullerenes on the intensity of colon damage and hematological signs of ulcerative colitis in rats. Mater Sci Eng C 1(93):505–517 Carter MJ, Lobo AJ, Travis SP (2004) Guidelines for the management of inflammatory bowel disease in adults. Gut 53(suppl 5):v1–6 Cho JY, Chang HJ, Lee SK, Kim HJ, Hwang JK, Chun HS (2007) Amelioration of dextran sulfate sodium-induced colitis in mice by oral administration of β-caryophyllene, a sesquiterpene. Life Sci 80(10):932–939 Ghasemi-Pirbaluti M, Motaghi E, Najafi A, Hosseini MJ (2017) The effect of theophylline on acetic acid induced ulcerative colitis in rats. Biomed Pharmacother 1(90):153–159 Hagar HH, El-Medany A, El-Eter E, Arafa M (2007) Ameliorative effect of pyrrolidine-dithiocarbamate on acetic acid-induced colitis in rats. Eur J Pharmacol 554(1):69–77 Kane S, Huo D, Aikens J, Hanauer S (2003) Medication nonadherence and the outcomes of patients with quiescent ulcerative colitis. Am J Med 114(1):39–43 Lennard-Jones JE (1989) Classification of inflammatory bowel disease. Scand J Gastroenterol 24(sup170):2–6 Morris GP, Beck PL, Herridge MS, Depew WT, Szewczuk MR, Wallace JL (1989) Hapten-induced model of chronic inflammation and ulceration in the rat colon. Gastroenterology 96(3):795–803 Mowat C, Cole A, Windsor AL, Ahmad T, Arnott I, Driscoll R, Mitton S, Orchard T, Rutter M, Younge L, Lees C (2011) Guidelines for the management of inflammatory bowel disease in adults. Gut 60(5):571–607
220
A. K. Netam et al.
Wallace JL, MacNaughton WK, Morris GP, Beck PL (1989) Inhibition of leukotriene synthesis markedly accelerates healing in a rat model of inflammatory bowel disease. Gastroenterology 96(1):29–36
Finite Element Analysis of Traumatic Brain Injury Due to Blunt Impact of Different Durations Tanu Khanuja and Harikrishnan Narayanan Unni
Abstract This study proposes a detailed biomechanical model of the human head to study the effect of non-penetrating (blunt) head impacts of different durations on intracranial organs. A patient-specific high biofidelity three-dimensional finite element human head model is developed from the magnetic resonance imaging (MRI) data and segmented into five volumes namely, skull, cerebrospinal fluid (CSF) with ventricular system, cerebrum, cerebellum, and brain stem and each segment is assigned with appropriate material properties. The model validated against the impact experiment based on human cadaver is used to perform simulation with a range of blunt impact durations and biomechanical analysis is performed by investigating the maximum intracranial pressure (ICP) and von Mises stress distribution across the brain. The probability of loss of consciousness and tissue damage is studied based on the ICP and von Mises stress values. The coup and contrecoup phenomena is also studied with the localization, extension and, intensity of tissue damage based on the injury tolerance criteria present in the literature. Keywords Blunt impact · Finite element human head model · Intracranial pressure · von Mises stress · Coup and ContreCoup phenomena
1 Introduction Traumatic brain injury (TBI) is the injury to the head organs due to sudden impact or jolt to the head which can result in temporary or permanent damage to the brain (Maas et al. 2008). Most of these injuries are invisible from the outside and current imaging modalities may or may not be able to detect the physical damage to the brain (Parikh et al. 2007). Typically, head impact durations causing TBI are in the range T. Khanuja (B) · H. N. Unni Department of Biomedical Engineering, Indian Institute of Technology, Hyderabad, India e-mail: [email protected] H. N. Unni e-mail: [email protected] © Springer Nature Singapore Pte Ltd. 2021 A. A. Rizvanov et al. (eds.), Advances in Biomedical Engineering and Technology, Lecture Notes in Bioengineering, https://doi.org/10.1007/978-981-15-6329-4_20
221
222
T. Khanuja and H. N. Unni
of few milliseconds and can cause closed or penetrating head injuries (Maas et al. 2008). However, it is not always possible to perform cadaveric experiments on the human brain to understand the injury phenomena. Consequently, the finite element method is used as a pronounced tool to understand the brain injury mechanism. The finite element method is used as a noninvasive approach to replicate injury phenomena to the head and to study the tolerance criteria for different types of injuries. In last three decades, a large number of numerical models of the human head are developed to study the computational mechanics of the skull fracture and TBI with linear material properties and simplistic geometries (Zhao et al. 2015; Siswanto and Hua 2012; Yan and Pangestu 2011; Yoganandan and Pintar 2003; Kleiven 2003). Recently, a few detailed geometrical structures modeled with nonlinear material properties are used for the computational study. However, these models have segmented the brain similar to a sphere or ellipsoid without the appropriate representation of sulcus and gyrus structures on the brain surface (Zhao et al. 2015; Yan and Pangestu 2011; Kleiven 2003). In addition, most of the experimental and computational studies focus on static and dynamic impact analysis with impact duration ranging from 3 to 20 ms (Nahum et al. 1977; Trosseille et al. 1992; El et al. 2008; Kleiven 2006). The short duration blunt impacts produce non-penetrating closed brain injury where the brain is not directly exposed to the impact (Hannay et al. 2004). These non-penetrating blunt impacts can be catastrophic and may not be detectable just after injury using imaging modalities. The computational study on patient-specific finite element head models can be used to give a clear picture of such blunt injury mechanisms. The present study focuses on the generation of a new three-dimensional finite element human head model from a patient-specific MRI data with a clear representation of sulci and gyri folds on the brain surface. In addition, the validated model is used for investigating the head injury mechanism with blunt impacts of durations ranging from 0.5 to 2.5 ms. The broad injury prediction and soft tissue damage criteria are discussed on the basis of threshold intracranial pressure (ICP) and von Mises stress summarized in literature.
2 Material and Methods 2.1 MRI Predicated 3-D Human Head Model and Mesh Generation At present, the magnetic resonance imaging (MRI) is the most sensitive imaging modality to accurately capture the soft tissue configurations of the human body (Lee and Newberg 2005). Hence, to specifically capture the sulci and gyri folding on the cerebrum surface the present finite element model is approximated from the axial MRI images obtained from “The whole brain atlas” of the Harvard medical school (Harvard Medical School). The image-based meshing platform AMIRA 5.6
Finite Element Analysis of Traumatic Brain Injury Due to Blunt Impact …
223
(Institute academic license) is used to generate the three-dimensional finite element human head model from the MRI image data. The image slices are labeled on the basis of region of interest (ROI) as represented in Fig. 1. The process of segmentation includes semi-automated tools such as thresholding, magic wand, contour fitting, and interpolation. In addition, the “remove islands” tool, “smooth labels” tool, and image processing filters are utilized to smoothen the segmented surface within the mask. The segmented surfaces are concatenated and converted to the coupled volumes. The segmented head model comprises of five rudimentary structures namely, skull, cerebrospinal fluid with ventricular system, cerebrum, cerebellum, and brain stem as illustrated in Fig. 2.
Fig. 1 Illustration of segmentation of ROI in one of the sagittal MRI slice
Fig. 2 Illustration of segmented volumes in AMIRA 5.6 a top view of cerebrum with sulci and gyri structures; b sagittal cross-section of full head surface model representing segmented layers; c front view of full transparent head model
224
T. Khanuja and H. N. Unni
Fig. 3 Illustration of the meshed model in AMIRA 5.6 with C3D4 type elements a cerebrum meshed with 188,742 number of elements; b sagittal cross-section illustrating different mesh segments; c full head meshed model with 345,597 number of elements
The three-dimensional finite element head model is meshed with 345,597 C3D4 (four nodes, constant strain, linear, solid tetrahedral elements with three degrees of freedom) elements and 62,568 nodes in AMIRA 5.6 using marching cube algorithm and grid-based meshing. In addition, the volumes are coupled using shared nodes between individual segments in order to avoid non-convergence of model simulation which is resulted from model complexity and contacts between structures. The individual segments skull, CSF, cerebrum, cerebellum, and brain stem are meshed with 60,462, 67,624, 188,742, 20,948, and 7821 number of C3D4 elements, respectively as illustrated in Fig. 3. Moreover, the mesh optimization is performed with varying mesh density of 103,890, 288,962, 345,597, and 579,352 elements in order to select the mesh model insensitive to the number of mesh elements. The maximum variation of 1.98% is observed in validation results in between models with 288,962, 345,597, and 579,352 elements. Therefore, to maintain the balance between accuracy and time cost, the model with 345,597 elements is utilized for further simulations.
2.2 Material Properties and Model Validation After meshing the model in AMIRA 5.6, the finite element head model is imported to the finite element solver ABAQUS 6.9/Explicit. Ideally, the skull is composed of three layers of different material properties and their own anisotropy. However due to lack of mechanical property data and complexity in layer segmentation, skull is modeled as a linear isotropic elastic material as represented in Table 1 (Siswanto and Hua 2012). The CSF is modeled as a solid linear elastic material with very small elastic modulus as illustrated in Table 1 (Yan and Pangestu 2011). However, being fluid with consistency similar to water, CSF should be modeled as Newtonian fluid but in order to avoid solid–fluid coupling difficulty during simulation CSF is modeled as a solid layer. The higher Poisson’s ratio is considered to represent the incompressibility of
Finite Element Analysis of Traumatic Brain Injury Due to Blunt Impact …
225
Table 1 Material properties for skull, CSF, and soft tissues Segment
Density (kg/m3 )
Elastic modulus (MPa)
Poisson’s ratio
Viscoelastic coefficients
Skull
3000
7300
0.22
–
CSF
1000
0.15
0.499
–
Brain (cerebrum, cerebellum, brain stem)
1040
0.0199
0.499
g1 = 0.5837, τ1 = 0.02571 g2 = 0.2387, τ2 = 0.0257
Table 2 Maximum ICP and von Mises stress during different impact pulses
Impact pulse duration (ms)
Maximum ICP (kPa)
Maximum von Mises stress (kPa)
2.5
183
69.75
2
206.185
64.4
1.5
226
55.8
1
237.96
43.2
0.5
476.5
22.9
the material. The brain tissue comprising of cerebrum, cerebellum, and brain stem are modeled as an incompressible linear viscoelastic material using two-term Prony series model as collated in Table 1 (Rashid et al. 2012; Zhang et al. 2001). The viscoelastic Prony series is given as: G(t) = 1 −
n
gi (1 − e
− τt
i
)
(1)
i=1
where, G(t) = Dynamic shear modulus, gi = relaxation coefficient, τ i = characteristics relaxation time. After assigning the material properties model is validated against the Nahum’s experimental data presented in literature (Nahum et al. 1977). The head-neck junction is considered to be freely moving and the semi-sinusoidal impact pulse of 6 ms duration similar to experimental contact force is applied to the frontal side of the skull. The computed ICPs at anterior and posterior sides of the model are compared with the experimental ICPs and a good correlation between model and experimental data is observed. The validation graph is presented by the author in the IEEE BIBE 2018 conference proceeding (Khanuja and Unni 2018).
226
T. Khanuja and H. N. Unni
3 Results and Discussion Subsequent to the model validation the blunt impact analysis is performed with impacts on frontal-top side of the head as shown in Fig. 4 with five impact pulses of semi-sinusoid shape ranging from 0.5 to 2.5 ms with peak magnitude of 9 kN as shown in Fig. 5. The magnitude of force is chosen based on the force range presented in literature to induce traumatic brain injury (Yoganandan et al. 1995).
Fig. 4 Illustration of impact location on the frontal-top side of the head
10
Force (kN)
8 6 4 2 0 0
0.5
1
1.5
2
2.5
time (ms) 2.5ms
2ms
1.5ms
Fig. 5 Illustration of impact force pulses applied on the head
1ms
0.5ms
Finite Element Analysis of Traumatic Brain Injury Due to Blunt Impact …
227
The analysis is performed in ABAQUS 6.9/Explicit with a minimum time increment of 70.9 ns for the 10 ms duration to analyze the immediate effect of impacts on the head. The maximum ICP and von Mises stress values for varying pulse durations are compared as presented in Table 2. It can be reported from the computed results that the maximum value of ICP is increasing with decreasing impact duration which is similar to the coup pressure trend presented in the literature with varying pulse durations (Pearce and Young 2014). In addition to this, our model predicts that the maximum von Mises stress is decreasing with decreasing impact duration. Based on the tolerance criteria existing in the literature the ICP exceeding 300 kPa causes mild traumatic brain injury while the von Mises stress exceeding 38 kPa generates 50% probability of severe neurological damage (Newman et al. 2000; Baumgartner and Willinger 2005). The ICP is used as an indicator of loss of consciousness and concussion and the von Mises stress is used as an indicator of neurological damage to the brain. Based on the threshold value of 300 kPa, the probability of loss of consciousness is observed to be maximum during the shortest impact pulse (0.5 ms) while the probability of maximum neurological damage is observed during longest impact pulse (2.5 ms) as illustrated in Fig. 6. Moreover, the maximum ICP is obtained near the peak of impact pulse while the maximum von Mises stress is obtained near 9 ms in all the cases. During impact to the frontal-top side of the head, a positive peak pressure is obtained on frontal lobe of the brain and on the other side, the negative pressure of similar magnitude range is observed in contrecoup side, covering occipital lobe of brain and cerebellum as illustrated in Fig. 7. The negative pressure may result in severe transient cavitation in the contrecoup side of the brain (El et al. 2008; Brennen 2003). In addition, after the removal of the impact, the negative pressure is also observed on the coup side which represents the probability of brain cavitation injury on coup side as well however of lesser severity. 2 1.8 1.6 1.4 1.2 1 0.8 0.6 0.4 0.2 0 0.5
1
ICP(simula on)/ICP(threshold)
1.5
2
2.5
von Mises (simula on)/von Mises (threshold)
Fig. 6 Probability of loss of consciousness (ICP ratio) and neurological damage (von Mises stress ratio) at 0.5, 1, 1.5, 2, and 2.5 ms duration impact pulses
228
T. Khanuja and H. N. Unni
Fig. 7 Illustration of coup and contrecoup ICPs during 2.5 ms duration impact pulse a positive pressure on coup site at 1.75 ms, b negative pressure on coup site at 4 ms
4 Conclusion The work presented the ICP and von Mises stress response of brain during short time pulse impacts of varying duration and same magnitude. For short duration impacts, the ICP has been seen as increasing with decreasing time width of the impact pulse whereas the von Mises stress is found to be decreasing with decreasing impact duration. The study concludes that the short duration impacts have higher probability to cause concussion and loss of consciousness without going through huge physical tissue damage. However, the increase in contact force duration will cause more damage to the intracranial organs with severe traumatic brain injuries. Furthermore, the coup and contrecoup pressure phenomena is studied for blunt impacts which indicate the probability of brain tissue cavitation injury on the coup site along with the contrecoup site of the brain but comparatively of less intensity at coup site. As the anatomical scale of the head varies based on the human head subjects, the present work can further be improved by analysis of the patient-specific finite element human head models of different age groups and genders. Acknowledgment The present work is supported by the Ministry of Human Resource and Development (MHRD), Government of India.
References Baumgartner D, Willinger R (2005) Numerical modeling of the human head under impact: new injury mechanisms and tolerance limits. In: IUTAM symposium on impact biomechanics: from fundamental insights to applications, pp 195–203 Brennen C (2003) Cavitation in biological and bioengineering contexts. In: Proceedings of the 5th international symposium on cavitation. Osaka, Japan
Finite Element Analysis of Traumatic Brain Injury Due to Blunt Impact …
229
El ST, Mota A, Fraternali F, Ortiz M (2008) Biomechanics of traumatic brain injury. Comput Methods Appl Mech Eng 197:4692–4701 Hannay HJ, Howieson DB, Loring DW, Fischer JS, Lezak MD (2004) Neuropathology for neuropsychologists. In: Lezak MD, Howieson DB, Loring DW (eds) Neuropsychological assessment. Oxford University Press, Oxford [Oxfordshire], pp 158–62. ISBN 978-0-19-511121–7 Johnson KA, Becker JA (2020) Normal Anatomy in 3-D with MRI/PET (Javascript), the whole brain atlas, Harvard Medical School. http://www.med.harvard.edu/AANLIB/cases/caseNA/pb9. htm. Accessed 31 July 2020 Khanuja T, Unni HN (2018) [Regular Paper] Computational modeling of traumatic brain injury due to impact on different sides of human head. In: IEEE 18th international conference on bioinformatics and bioengineering (BIBE), Taichung, Taiwan, pp 364–370 Kleiven S (2003) Influence of impact direction on the human head in prediction of subdural hematoma. J Neurotrauma 20(4):365–379 Kleiven S (2006) Evaluation of head injury criteria using a finite element model validated against experiments on localized brain motion, intracerebral acceleration, and intracranial pressure. Int J Crashworthiness 11:65–79 Lee B, Newberg A (2005) Neuroimaging in traumatic brain imaging. NeuroRx 2(2):372–383 Maas AI, Stocchetti N, Bullock R (2008) Moderate and severe traumatic brain injury in adults. Lancet Neurol 7(8):728–741 Nahum AM, Smith RW, Ward CC (1977) Intracranial pressure dynamics during head impact. In: Proceedings 21st stapp car crash conference, pp 339–366 Newman J, Barr C, Beusenberg M, Fournier E, Shewchenko N, Welbourne E, Withnall C (2000) A new biomechanical assessment of mild traumatic brain injury, II: results and conclusions. In: Proceedings of international IRCOBI conference on the biomechanics of impact, pp 223–233 Parikh S, Koch M, Narayan RK (2007) Traumatic brain injury. Int Anesthesiol Clin 45(3):119–135 Pearce CW, Young PG (2014) On the pressure response in the brain due to short duration blunt impacts. PLoS ONE 9(12):e114292 Rashid B, Destrade M, Gilchrist M (2012) Hyperelastic and viscoelastic properties of brain tissue in tension. In: Proceedings of the ASME 2012 international mechanical engineering congress and exposition, November 9–15, 2012, Houston, Texas, USA Siswanto WA, Hua CS (2012) Strength analysis of human skull on high speed impact. Int Rev Mech Eng 6(7):1508–1514 Trosseille X, Tarriere C, Lavaste F, Guillon F, Domont A (1992) Development of a F.E.M. of the human head according to a specific test protocol. In: Proceedings on 36th stapp car crash conference, SAE Paper No. 922527, Society of Automotive Engineers, Warrendale, PA Yan W, Pangestu OD (2011) A modified human head model for the study of impact head injury. Comput Methods Biomech Biomed 14(12) Yoganandan N, Pintar FA (2003) Biomechanics of temporo-parietal skull fracture. Journal of Clinical Biomechanics 19(3):225–239 Yoganandan N, Pintar FA, Sances A, Walsh PR, Ewing CL, Thomas DJ, Snyder RG (1995) Biomechanics of skull fracture. J Neurotrauma 12(4):658–668 Zhang L, Yang KH, King AI (2001) Comparison of brain responses between frontal and lateral impacts by finite element modeling. J Neurotrauma 18(1):21–30 Zhao W, Ruan S, Ji S (2015) Brain pressure responses in translational head impact: a dimensional analysis and a further computational study. Biomech Model Mechanobiol 14(4):753–766
Data Dissemination Using Social-Based Attributes in Delay-Tolerant Networks Sanjay Kumar, Prasoon Shukla, and Sudhakar Pandey
Abstract In rural areas an ambulance carrying a patient may require to connect with experts in a city. In rural areas it is not possible that network connectivity to the ambulance available all the time. In this type of situation concept of delaytolerant networks may be utilized for the opportunistic connectivity with a person at distance. Delay-Tolerant Network (DTN) is an approach to a wireless network, which possesses an uncertainty in the connection. Due to intermittent connectivity and long delays, routing in DTN turns out to be quite challenging. In recent years, many socialbased routing reflects the capabilities of social attributes for disseminating data. In this paper social attributes namely, community, similarity, betweenness, and degree centrality are being used for outlining a routing technique. Using this approach we can improve the quality of service provided by the ambulances while passing through rural areas. A comparison is made with already proposed routing technique, for example, BUBBLE Rap, Epidemic, and experiences a noteworthy improvement in delivery ratio and buffer utilization. Keywords Delay-tolerant network · Community · Degree centrality · Betweenness centrality · Similarity
1 Introduction Rural areas do not provide network connectivity to the vehicles passing through these areas due to lack of infrastructure. Vehicles carrying patients may require to establish connection with the doctors available in nearby cities. But this facility can be added to the ambulances only when there is some mechanism to provide network connectivity in the areas where end to end network connectivity is not available all the time. Delaytolerant networks can be one of the solutions in this type of scenario. Delay-tolerant network (DTN) (Fall 2003; Delay Tolerant Networking Research Group 2004) is an opportunistic network where the connection is intermittent, i.e., network is sparse in S. Kumar (B) · P. Shukla · S. Pandey Department of Information Technology, National Institute of Technology, Raipur, India e-mail: [email protected] © Springer Nature Singapore Pte Ltd. 2021 A. A. Rizvanov et al. (eds.), Advances in Biomedical Engineering and Technology, Lecture Notes in Bioengineering, https://doi.org/10.1007/978-981-15-6329-4_21
231
232
S. Kumar et al.
nature. Lack of infrastructure, partitioning of the network, very long delays are the features of delay-tolerant network. Due to these features routing in such network is a quite challenging task and numerous researches are going on. Routing in delaytolerant network concerns itself with the ability to route or transport the data packets (contains message) from a source to destination It is a generally a difficult task to design an efficient routing protocol for opportunistic networks due to the lack of knowledge about the infrastructure of the network and dedicated connection is not present always present. Routing (Balaganesh et al. 2014; Zhu et al. 2013) in the delay-tolerant network is mainly characterized into two types, namely Flooding- and Forwarding-based strategy. In Flooding-based routing (Balaganesh et al. 2014; Zhu et al. 2013), the multiple copies of the same message will be created and these copies will be delivered to all set of nodes that are encountered by the node containing the data packet. The process goes on till the message is delivered to the destination or time to live (TTL) ends. Some Flooding-based routing techniques are Epidemic routing, Spray, Wait, Single and Two Hop routing techniques, etc. Whereas Forwarding-based (Balaganesh et al. 2014; Zhu et al. 2013) routing uses the knowledge of the network (such as network topology) for determining the relevant path to deliver a message to the destination. Some forwarding-based routing techniques are Source routing, PerHop, and Per-contact routing. One special category of the routing techniques exists, namely, Social-based routing (Balaganesh et al. 2014; Zhu et al. 2013) where social features of an opportunistic network are taken to route the message. Some of the social-based routing techniques are Location-Based routing, Label-Based routing, BUBBLE RAP routing, etc. In this paper, the main intention is to exploit social behaviors of the delay-tolerant network with the help of social attributes: community, similarity, betweenness, and degree centrality. A community is a social unit (a group of people) formed on the basis of some common features such as profession, needs, place (the geographical area where they reside). Within a community (social institutions like family, government, society, etc.) everyone has some role associated with them due to which some interact with more number of peoples while some don’t even interact. On this basis popularity of a particular human being is determined, termed as centrality. Every individual belongs to one or more communities and has some likeness with other people. The degree of likeness is measured by similarity which describes how often people interact and pattern they follow to interact with others. In this paper, community, similarity, betweenness, and degree centrality are being exploited for forwarding data packets in DTN. Methodologically, community detection helps in understanding local community structure which further used in designing efficient routing techniques for disseminating data. Newman (2004) has given many centralities to approximate the importance of a node in a network. Betweenness centrality is counted as the frequency a given node encountered on the shortest path followed between a pair of nodes. Degree centrality of a particular node measures the number of nodes encountered to that given node. Similarity measures the likeness between two given nodes by estimating the number of common neighbors between these given nodes. The combination of these three helps in selecting the best possible node to forward the data packets.
Data Dissemination Using Social-Based Attributes …
233
Improvement in the delivery probability and buffer utilization with the help of these social attributes are contributions of the paper we are presenting. The rest of the paper is structured as follows: Sect. 2, 3, 4, and 5 explains the social attributes and their significance in routing the data packets. Section 6 introduces the proposed routing technique containing these social attributes in selecting the route and their roles in proposed routing technique. Section 7 deals with performance evaluation of the proposed routing technique in terms of delivery probability and buffer utilization. Section 8 presents the conclusion.
2 Community A community (George 1955; Gondaliya et al. 2015; Danon et al. 2005; Gor and Dhamecha 2014) is a subgraph where intra-cluster density is greater than inter-cluster density when the density of edges is taken into an account. In a generalized way, when things, individuals, creatures are gathered together in view of some vicinity (e.g., space, time, intrigue, and so on) then they frame a community (e.g., Hindu community and Muslim community, IT Professional community, and CIVIL professional community portrayed in light of religion and calling individually). It is one of the prime attributes of delay-tolerant networks (DTN) as mobile devices are carried by a human being who belongs to certain communities. In BUBBLE RAP routing technique, it was expected that every last node is at least in one community whether it is global or subcommunity. The global community is that in which each and every node exist that are present in the network. Whereas, in subcommunity, every one of those nodes is available which exist in the global community and are grouped together with a view of some social elements. In this proposed routing technique, the same suspicion is additionally taken. The various community detection algorithms have been proposed to identify communities in a network such as Weighted Network Analysis (WNA) by Newman (2005) and K-CLIQUE proposed by Palla et al. (2005) which helps in finding communities in a network. In this paper for community detection (CONGA), Cluster-Overlap Newman–Girvan Algorithm (Gregory 2007; Jora and Chira 2016; Kim et al. 2013; Khatoon and Banu 2015; Jancura 2012) is being used, which is an enhanced version of Grivan and Newman (GN) algorithm (Gregory 2007). GN algorithm is first divisive detection algorithm which espouses betweenness to extricate communities from a number of nodes but suffers from poor scalability. CONGA came up with a new concept of “splitting” betweenness of vertices, which enables it to split vertices among various communities. In order to split a vertex V into V1 and V2, firstly a new “imaginary” edge is being added. The cost of the imaginary edge is taken zero. Moreover, no possible paths originating from vertex V traverse this edge hence; the path lengths traversing V will remain untouched and no new shortest paths will be created. Then betweenness CB{V1, V2} needs to be calculated for imaginary edge. V can be splited into two parts choosing one from the 2d (v)-1-1 ways.
234
S. Kumar et al.
3 Betweenness Centrality In graph theory, centrality (Kim et al. 2014; Freeman 1977; Wan et al. 2016) is a quantitative measure of the relative significance of a vertex within a graph. In the social graph, the centrality of a node delineates its social significance within the social network (e.g., how important a person is within a social network and reflects his activeness). In DTN, the sociological centrality is being used for selecting a node as a relay for forwarding the messages, which means nodes with higher centrality are the best nodes to be selected as a relay. Betweenness centrality (Freeman 1977; Newman 2005; Manam et al. 2014) of a node “x” is defined as the number of times node “x” is encountered as a relay in all possible shortest paths between two other nodes (Fig. 1). It is one of the important social attributes as it describes the amount of data disseminated through it. Let us take a graph G = (V, E), where V and E represent the set of nodes and edges in sparse network respectively. “a” and “b” are the numbered nodes and edges, respectively. It is assumed that each edge e ∈ E has a positive integer weight w(e) = 1 which shows that there exists a dedicated connection between the nodes. A path from vertex source (s) to destination (d) is defined as a sequence of vertices vi , 0 ≤ i < l, where v0 = s and vl = d. A path length is defined as the sum of the weights of edges. w=
l
w(ei )
∀ vi , 0 < i < l
(1)
i=0
σ (s,d) is the total shortest path between sand d, and σ (s,v,d) represents the number of paths passes through v. τ (s,v,d) we take as the fraction of shortest paths between s and d passing vertex v, i.e., τ (s, v, d) = σ (s, v, d) σ (s, d)
(2)
Betweenness centrality C B (v) of a vertex v is defined as Fig. 1 Between centrality of a node taking 1 as source and 100 as destination in 100 node sparse network
1
0.6 0.4 0.2
Node_ID
97
89
81
65
73
57
49
33
41
25
17
9
0 1
CB'(v)
0.8
Data Dissemination Using Social-Based Attributes …
C B (v) =
235
τ (s, v, d)
(3)
s,d∈G(V,E)
As the networks are exceedingly large so calculating the betweenness indices needs polynomial running time complexities of O(n2 ) or O(n3 ) and are still prohibitive for networks as it contains a large number of nodes. Moreover, the networks are updated continuously, adding or removing node being able to track the changes in a network and the accompanying centrality is quite tough and infeasible. The local approach uses only the vertices directly adjacent to a target vertex to derive an approximation of the true centrality measure. In order to obtain a local approximation for betweenness centrality, the numerator in (2) can be decomposed as σ (s, v, d) = σ (s, v) · σ (x, d)
(4)
Putting the value of σ in (2) and then combining it with (3) we move toward the approximation of CB. After repeatedly applying the decomposition identity as done in (2) lead to those fractions which only includes a predecessor and a successor of v0 , v1 represents predecessor and successor of the node v. Thus, C B (v) =
N N τ (v0 , v, v1 ) a=1 b=1
τ (v0 , v1 )
(5)
4 Degree Centrality Degree centrality (Kim et al. 2014; Wu et al. 2016; Wan et al. 2016) of a node is determined by the number of links present in that node means a number of nodes connected to it. A node having more number of connections possesses more possibility to deliver a message successfully and can be considered as a popular node. Let the network contains N number of edges then degree centrality C D (x) of node “x” is calculated by: C D (x) =
N
σ (x, i)
(6)
i=1
where σ (x, i) represents the status of the connection between node “x” and “i.” The value of σ (x, i) will be either 1 or 0. “1” indicates that there exists a connection between the nodes whereas the opposite holds true for “0.” Now approximating the degree centrality with respect to the maximum possible degree of a node.
236
S. Kumar et al.
Fig. 2 Degree centrality of a node taking 1 as source and 100 as destination in 100 node sparse network
1
CD'(x)
0.8 0.6 0.4 0.2 89
97
81
65
73
57
49
41
33
25
17
9
1
0
Node_ID
C D (x) =
N σ (x, i) i=1
ρ(x)
(7)
ρ(x) represents the maximum possible degree of the node in the network. If the network contains N number of nodes then the maximum possible degree of a node will be ρ(x) = N − 1. So. Putting The Value of ρ(x) in (7), we get C D (x)
=
N σ (x, i) i=1
N −1
(8)
In this proposed routing technique, both degree and betweenness centrality are being used. They are used to define the significance of a node in a given network which further used to select the best possible node for forwarding the data in that network. The graph below in Fig. 2 shows the degree centrality of various nodes at a particular instance of time determined from the sample network of 100 nodes taken below.
5 Similarity Similarity (Zhu et al. 2013; Daly and Haahr 2007; Patel and Gondaliya 2015) is a sort of social attributes which is used to express the relation among the nodes in terms of location, interest, background, etc. It is measured based on common neighbors present between the pair of nodes in a given community whether it is global or subcommunity. Similarity came from the observation that individuals often befriend others who have a similar appeal and perform a similar activity. In this manner, the high similarity between the nodes infers the presence of a good social connection between them. The probability of a future collaboration P(a,b) can be anticipated by the result of similarity and its strength between two nodes “a” and “b” is calculated
Data Dissemination Using Social-Based Attributes …
237
by: P(a, b) = ρ × S(a, b) S(a, b) =
n
(9)
|M(a) ∩ M(b) |
(10)
i=1 j=1
S(a, b) =
n
|M(i, j) |
(11)
i=1 j=1
where M(a) and M(b) are the set of neighbors of node “a” and “b”, respectively. This S(a,b) reveals the “similarity” between nodes “a” and “b,” relative to the network topology. The strength of a similarity is defined as the number of times a pair of nodes interacts with each other in a given period of time. Let us suppose in a network 8 numbers of nodes are present and at a particular instance of time then the connection matrix be: a
B
c
d
e
f
g
h
a
0
0
1
0
1
1
0
1
b
0
0
1
1
0
1
0
1
c
1
1
0
0
0
1
0
1
d
0
1
0
0
1
0
0
0
e
1
0
0
1
0
1
1
0
f
1
1
1
0
1
0
1
0
g
0
0
0
0
1
1
0
1
h
1
1
1
0
0
0
1
0
If in a network “N” number of nodes are present then the size of connection matrix be “N × N.” Here in above the size of connection matrix “N” is 8, which represent a number of nodes and connection between them. The value of connection matrix changes with time as each is associated with its mobility pattern which changes with time. Then the value of M(a) and M(b) can be calculated from connection matrix. M(a) matrix is given by: 0
0
1
0
1
1
0
1
1
0
1
0
1
M(b) matrix is given by: 0
0
1
Resultant matrix M is given by:
238
0
S. Kumar et al.
0
1
0
0
1
0
1
Therefore similarity between the nodes a and b will be: S(a,b) = 3. The strength of the similarity (ρ) between the particular pair of nodes will be calculated by the contact history of the pair over a period of time. In the social graph, strength can be represented as the number of times two human beings (belonging to the same community) interact with each other over a specific period. Let us suppose x(t) be a function which represents the interaction pattern between a given pair of nodes. Mathematically, it is calculated as the integration of the total area of all instances of function x(t) over a period “T ” (when it is active with another node) divided by area of one instance of a function x(t) calculated over the same time period. Then the value of ρ is given by: T ρ = t=0 t1 t=0
x(t) x(t)
(12)
where the value of x(t) will be 1 for a traverse of time when two nodes are in contact with each other else it will be 0.
6 Proposed Algorithm The selection procedure of the node follows in two major steps. These steps are as follows: Step 1: Each node has its centrality (both betweenness and degree) associated with it. The source node will select the best possible node among its all encountered node that is having the highest betweenness centrality. Then it will be selected as a relay node. On the off chance that more than one node is having the same value of betweenness centrality then next level of selection of a node is done in order to select best possible node. Now degree centrality is mulled over for choosing the most appropriate node among those nodes which is having same betweenness centrality. This procedure of node selection is done till message reaches the destination community. As soon as it reaches the destination community step 2 is followed to perform further forwarding of the message. Step 2: As the message reaches the destination community then a blend of two social attributes, to be specific degree centrality and the similarity are utilized to select the node for forwarding the message. These attributes are utilized within the destination community as opposed to in a global community because intra-community density is higher than that of inter-community density. Since these social attributes are determined from quite different criteria so a utility function is shaped to consolidating these social attributes, which will be utilized for the decision-making process.
Data Dissemination Using Social-Based Attributes …
239
U (s, x, d) = α × P(s, x) + β × C D (x)
(13)
Now, U (s, x, d) = α × ρ ×
n
|M(i, j) | + β ×
i=1 j=1
n σ (x, j) n−1 j=1
(14)
The value of α and β is chosen such a way that relative significance of the two social-based attributes can be perfectly utilized and α + β = 1. Among all encountered nodes that node will be selected which is having the highest value of the utility function and this process continues until the destination node is reached. Since the buffer size is fixed and if not legitimately oversaw then leads to overhead in the network. So for proficient utilization of buffer various buffer management techniques were proposed (Le et al. 2016; Rashid et al. 2015). Here a message having most minimal TTL will be dropped out when the buffer of the node selected for forwarding the message is full. The message with least TTL is dropped with a view that the probability of successful delivery of that message is less than that of another message.
7 Performance Evaluation For evaluation-proposed routing technique is evaluated and compared against three existing DTN routing techniques, namely Epidemic routing, BUBBLE Rap, Direct Delivery. The evaluation of these routing techniques is done in a scenario that node generates messages in a specific interval of time during the simulation. Parameters for simulation setups are shown in Table 1.
7.1 Buffer Utilization Space (memory) assumes a critical part in the routing technique as buffer size of the node is limited and can’t be expanded because of different issues, for example, administration issue and power utilizations. Additionally, if irrelevant messages are Table 1 Simulation Setup for evaluation of routing techniques
Simulation
Setup
Number of nodes
80
Message TTL
300 min
Message sizes
10–100 KB
Buffer sizes
1–10 MB
240 50
No. of data packets
Fig. 3 Number of data packets used for the successful delivery of message
S. Kumar et al.
Epidemic
40 BUBBLE Rap
30 20
Proposed
10 0 10
32
53
66
Direct Delivery
Destination Node ID
put away then it leads to the dropping of numerous message bundles. So messages should be stored in a productive way and oversee them. In various multi copies routing techniques such as the Epidemic routing extensive number of the message, imitations are being made, and a number of copies rely on a number of encountered nodes. In the proposed routing technique, two levels of selection are done to choose one most appropriate node for forwarding the message. Therefore, it leads to a significant decrement in the usage of the buffer. The space complexity of a routing technique is defined as the number of replicas of a message created for successful delivery of the single message. As in BUBBLE Rap and Epidemic routing techniques, multiple paths are chosen for forwarding the message whereas in the proposed routing technique single most relevant path is chosen. So, a number of replicas will be less which is shown in Fig. 3 where distinctive destination nodes are chosen randomly with fixed source node “1” in a dynamic network. From Fig. 3 it can be presumed that for same destination number of replicas utilized by proposed routing is relatively not as much as that of existing routing techniques. For the fixed buffer size there is a significant dropping of messages in Epidemic routing and BUBBLE Rap. So for efficient utilization of buffer message having most minimal TTL will be dropped out when the buffer of the node selected for forwarding the message is full as explained in Sect. 6. In Fig. 4, each result in this section is the average result of 50 simulations for 1 MB buffer size scenario, proposed routing technique has higher Buffer Utilization Ratio (BUR) than BUBBLE Rap by 27.3% and Epidemic routing by 37.5% because of optimal node and path selection.
7.2 Delivery Ratio The delivery ratio is defined as the ratio of a number of messages received by the destination node to those generated by the source node. Suppose a number of messages generated by the source node “s” destined to destination node “d” be “M S ” and
Fig. 4 Relative buffer uutilization ratio of overall network for various destination nodes
Buffer Utilization Ratio
Data Dissemination Using Social-Based Attributes …
241
0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 10
32
Epidemic
53
66
BUBBLE Rap
Proposed
messages received by the destination node be “M D .” Mathematically, it can be defined as: Deliver y Ratio = M D M S
(15)
Fig. 5 Buffer utilization ratio of overall network for various destination nodes
Delivery Ratio
In Fig. 5 routing technique is compared with proposed technique and for various times to live of a message generated by the source node. The delivery ratio is been approximated by the simulation run of 50 times. From Fig. 5, it can be concluded that the proposed routing technique possess delivery ratio greater than BUBBLE Rap and Directly delivery but lower than Epidemic routing technique. 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 100
200
300
400
500
TTL (minutes) Epidemic
BUBBLE RAP
Proposed
Deirect Delivery
242
S. Kumar et al.
8 Conclusion The conventional wireless networks have been quite questionable due to its long delays and sparse nature. However, social-based routing supplemented with the social traits like community, centrality, and the similarity is a superior answer for the conventional delay-tolerant networks. In this paper by utilizing social attributes, namely community, similarity, betweenness, and degree centrality an optimal selection of the node for forwarding the data packets. The proposed routing technique has been compared with existing routing techniques such as BUBBLE Rap, Epidemic routing, and direct delivery technique. After comparisons, we thought of a determination that there is a noteworthy change in the delivery ratio and buffer has been efficiently utilized. This prompts significant improvement in routing more and more data packets at once. Finally this approach may be implemented in real-world situation like a ambulance carrying a patient through rural areas and communication with doctor in hospital is not regular. The proposed technique will be helpful in improving the quality of service of communication between a doctor and ambulance. Proposal has not been tested in real-world scenario which limits its applicability.
References Balaganesh M, Sathiya P, Balagowri D (2014) A survey of misbehavior detection scheme in DTN. Int J Comput Appl 107(10) Daly EM, Haahr M (2007) Social network analysis for routing in disconnected delay-tolerant MANETs. In: Proceedings of the 8th ACM international symposium on mobile ad hoc networking and computing, pp 32–40. ACM Danon L, Diaz-Guilera A, Duch J, Arenas A (2005) Comparing community structure identification. J Stat Mech: Theory Exp 2005(9):P09008 Fall K (2003) A delay-tolerant network architecture for challenged internets. In: Proceedings of the 2003 conference on applications, technologies, architectures, and protocols for computer communications, pp 27–34. ACM Freeman LC (1977) A set of measures of centrality based on betweenness. Sociometry, 35–41 George Jr A (1955) Definitions of community: areas of agrement rural sociology. Hillery 20:118–118 Gondaliya N, Shah M, Kathiriya D (2015) A node scheduling approach in community based routing in social Delay Tolerant Networks. In: 2015 International conference on advances in computing, communications and informatics (ICACCI), pp 594–600. IEEE. Gor HR, Dhamecha MV (2014) A survey on community detection in weighted social network. Int J 2(1) Gregory S (2007) An algorithm to find overlapping community structure in networks. In: European conference on principles of data mining and knowledge discovery, pp 91–102. Springer, Berlin, Heidelberg Jancura P (2012) Evolutionary analysis in PPI networks and applications. [Sl: sn]. Jain S, Fall K, Patra R (2004) Routing in a delay tolerant network. In: Proceedings of the 2004 conference on applications, technologies, architectures, and protocols for computer communications, pp. 145–158 Jora C, Chira C (2016) Evolutionary community detection in complex and dynamic networks. In: 2016 IEEE 12th international conference on Intelligent computer communication and processing (ICCP), pp 127–134. IEEE
Data Dissemination Using Social-Based Attributes …
243
Khatoon M, Banu WA (2015) A survey on community detection methods in social networks. Int J Educ Manage Eng (IJEME) 5(1):8 Kim CM, Kang IS, Han YH, Park CY (2013) A community detection scheme in delay-tolerant networks. In: Ubiquitous information technologies and applications, pp 745–751. Springer, Dordrecht Kim CM, Kang IS, Han YH, Jeong YS (2014) An efficient routing scheme based on social relations in delay-tolerant networks. In: Ubiquitous information technologies and applications, pp 533–540. Springer, Berlin, Heidelberg Le, T., Kalantarian, H. and Gerla, M., 2016, June. A joint relay selection and buffer management scheme for delivery rate optimization in dtns. In World of Wireless, Mobile and Multimedia Networks (WoWMoM), 2016 IEEE 17th International Symposium on A (pp. 1–9). IEEE. Manam VC, Mahendran V, Murthy CSR (2014) Performance modeling of DTN routing with heterogeneous and selfish nodes. Wireless Netw 20(1):25–40 Newman ME (2004) Analysis of weighted networks. Phys Rev E 70(5):056131 Newman ME (2005) A measure of betweenness centrality based on random walks. Soc Networ 27(1):39–54 Palla G, Derényi I, Farkas I, Vicsek T (2005) Uncovering the overlapping community structure of complex networks in nature and society. Nature 435(7043):814 Patel CM, Gondaliya N (2015) Enhancement of social based routing protocol in delay tolerant networks. Int J Comput Appl 122(4) Rashid S, Ayub Q, Abdullah AH (2015) Reactive weight based buffer management policy for DTN routing protocols. Wireless Pers Commun 80(3):993–1010 Wan L, Zhang H, Liu F, Chen Y (2016) Routing in Delay tolerant networks with fine-grained contact characterisation and dynamic message replication. In: 2016 IEEE 17th international symposium on a world of wireless, mobile and multimedia networks (WoWMoM), pp 1–6. IEEE Wu J, Wang J, Liu L, Tanha M, Pan J (2016) A data forwarding scheme with reachable probability centrality in DTNs. In: 2016 IEEE on Wireless communications and networking conference (WCNC), pp 1–6. IEEE Zhu Y, Xu B, Shi X, Wang Y (2013) A survey of social-based routing in delay tolerant networks: Positive and negative social effects. IEEE Commun Surv Tutor 15(1):387–401
Hiding Patient Information in Medical Images: A Robust Watermarking Algorithm for Healthcare System Ritu Agrawal, Manisha Sharma, and Bikesh Kumar Singh
Abstract Electronically transmission of the medical images is an essential prerequisite in a Healthcare system. During transmission, the medical images can be hacked and altered either partially or in totality, unless aptly safeguarded. Digital watermarking schemes are being utilized to protect the integrity of the medical image. Digital watermarking is a technique of embedding watermark in a medical image, to maintain confidentiality and security to the inserted watermark. The watermark used in the medical image can either be an image, text, audio, or video. Electronic Patient Record (EPR) in an image form is used as a watermark. A non-blind watermarking scheme for healthcare system using Discrete Cosine Transform (DCT) is presented in this paper. Block-based Cosine transformation is applied to the host medical image to obtain different frequency coefficients. The middle frequency band of Cosine transformation is considered for embedding watermark, as this frequency band provides an extra robust resistant to watermark. Before hiding the EPR information in the image, binary EPR information is encoded using Convolution Error Correcting Code (ECC) and decoded using Viterbi decoder, to enhance the accuracy of the detection process. To provide additional robust to the coded EPR data, M-ary modulation is further applied before embedding. The imperceptibility and robust performance evaluation of the scheme is tested on standard online Digital Imaging and Communications in Medicine (DICOM) brain image database by changing the watermark embedding factor. The proposed scheme is tested using various performance measures and is observed to be highly robust and imperceptible. Keywords Medical image watermarking · Electronic patient record · Error control coding and M-ary modulation R. Agrawal (B) · M. Sharma Department of Electronics and Telecommunications, Bhilai Institute of Technology, Durg, Chhattisgarh, India e-mail: [email protected] B. K. Singh Department of Biomedical Engineering, National Institute of Technology, Raipur, Chhattisgarh, India © Springer Nature Singapore Pte Ltd. 2021 A. A. Rizvanov et al. (eds.), Advances in Biomedical Engineering and Technology, Lecture Notes in Bioengineering, https://doi.org/10.1007/978-981-15-6329-4_22
245
246
R. Agrawal et al.
1 Introduction Maintaining health is a prerequisite for sustaining the human resources, the most valuable resources available on the earth. Active assistance of physicians and associated personnel are required to achieve better health. In the age of converging technologies, providing information about a patient for correct and accurate diagnosis through electronic means and media plays a vital role. Due to rapid growth in the domain of Information and Communication Technology (ICT), many advanced means like telemedicine, telesurgery, telediagnosis, etc., have been evolved to ease the workings of the physicians with the active usage of internet and multimedia technologies. However, the privacy of medical data in electronic form poses potential threats both external as well as internal to maintain the patient–physician relationship for accurate delivery of healthcare. Internal threats can be overcome to a greater extent through proper management whereas external threats are to be handled by exploiting the existing technologies. For efficient and effective implementation of the patient– physician relationship, healthcare systems, it is necessary to share medical information over open networks. Exchange of information over open network is raising various complex legal and ethical issues, including image retention and fraud, privacy, malpractice liability, etc. Thus the major challenge to maintain a better patient–physician relationship, privacy and secured transaction of medical images is of prime importance. As the intruders pose a challenge to the privacy of the medical images being exchanged over an open communication channel, it needs to be addressed and resolved. The issue of unwanted access during transmission of the image can be effectively handled by using Hiding technique. Hiding is a technique in which the message signal or the information embed in the host/cover images without any much degradation in the visual quality to the host image. There are two major techniques of hiding the image during transmission, namely, Steganography and Watermarking schemes. In this paper we only address the watermarking scheme in the medical image. Watermarking of medical images addresses the specific issues of maintaining the integrity of medical images that provides the assurance that the data was not accidentally or deliberately modified during transmission over communication channels. Medical images are as important as other medical data. They play a key role in almost all phases of the healthcare systems like detection, diagnosis, therapy follow-up, and so on. Hence hiding of the image data during transmission is of major concern. The present work deals with the designing of a novel watermarking scheme in the area of medical imaging. In the watermarking scheme, the DCT as a host image is used for insertion and retrieval of the watermark. A combination of ECC and M-ary modulation was applied to the watermark, after that the coded message was inserted in the mid-band DCT coefficient of the cover medical image. Choice of considering mid-frequency Cosine transformation band was to provide good imperceptibility and robust to the inserted watermark.
Hiding Patient Information in Medical Images …
247
Related work A chronological survey on the previous works made by the researchers in the field of medical image watermarking schemes and embedding patients’ information in the cover image is presented in the following paragraphs. However looking into the vastness of the field undertaken, the survey has been presented in parts like information hiding and enhancement of robustness of the system using spread spectrum technique. Information Hiding. Mostafa et al. (2010) formulated a blind EPR hiding scheme using Discrete Wavelet Packets Transform (DWPT) in the application of telemedicine. Before embedding watermark, BCH ECC was applied to the binary watermark, for improving the robust property of the watermarking scheme. Though, the watermark embedding scheme is simple but the time required for extracting the watermark was high due to BCH error correction code. Nambakhsh et al. (2011) proposed a contextual based dual watermarking scheme for PET images. Two watermarks, text information and ECG signal of the patient were inserted in the selected preferred texture area of the medical image. The robust assessment of the scheme was tested only for three known attacks. The security of the system was greatlhanced as dual watermarks were embedded in the PET image but the system required higher execution time due to higher level of complexity of the system. Das and Kundu (2011) developed a multiple and non-blind watermarking scheme using contourlet transform and Discrete Cosine transform. The perceptual quality and the robust performance of the retrieved watermark were high as combinations of transform domain were used. The system was found to be complex as two different transformation techniques were used. Kannammal and Shuba Rani (2014) reported a dual-level security for medical image transaction using encryption and watermarking. The watermarking techniques used were LSB and Wavelet transform. Before embedding the watermark, different types of encryption technique like RSA, AES, and RC4 were applied to provide security to the watermark image. Due to the application of different encryption techniques for embedding, the system became very complex and the execution time was high. Singh et al. (2015a) reported a robust, non-blind, and multiple watermark embedding technique using Singular value decomposition (SVD) and Discrete Wavelet Transform (DWT) for telemedicine application. Moreover, to provide additional robustness to text watermark four different ECC were applied. Owing to the application of dual watermarks, the security scheme got remarkably improved, however, due to system complexity the execution time requirement also got enhanced. Singh et al. (2015b) devised a novel spread spectrum based multiple watermark embedding scheme using DWT. To improve the robust performance of the test watermark error control coding BCH was applied. Though the security of the watermarking system got enhanced as spread spectrum technique is applied yet the time required for extracting the watermark was high due to BCH error correction code and the
248
R. Agrawal et al.
system’s complexity increased due to application of two different transformation techniques. Mehto and Mehra (2016) devised a combined watermarking algorithm using DWT and DCT. EPR watermark was inserted in the host image such that the technique is imperceptible to the Human Visual System (HVS). The watermarking technique is too complex as two different transformation techniques were used for the embedding and extraction process. Moreover, no common processing attacks were applied to evaluate the robustness of the system. Swaraja (2017) reported a multiple watermark embedding scheme using DCT, DWT, and SVD for authentication and integration in the application of medical images. The security of the watermarks was enhanced by applying lossless arithmetic coding to text watermark and Arnold transform to the image watermark. On the contrary, the system became complex as different transformation techniques were used for the embedding and extraction process. Also, no common processing attacks were applied to evaluate the robustness of the system. Zear et al. (2018) formulated a combined watermarking scheme in the application of Healthcare using DCT, DWT, and SVD. To enhance the security of the multiple watermarks were applied. The proposed scheme performances were examined using different parameters at varying gain factor. Use of the multiple watermarks the security of the system increased, however, the higher complexity of the system increased to execution time. Enhancement of robustness of watermarking system using spread spectrum techniques. In digital watermarking the Spread Spectrum (SS) modulation principle has been applied due to its excellent characteristics viz. security protection and robustness. The use of the wide spectrum of the host signal in the message hiding process puts a limit on the data rate subject to a given embedding distortion. Kutter (1999) reported that the role of M-ary modulation is an effectual way to get a substantial improvement in robust performance. Also the performance did not diminish under noise like distortion, for example, lossy JPEG compression. Hajjaji et al. (2011) designed a secure based watermarking scheme for transmission of medical images using a hybrid combination of Code Division Multiple Access (CDMA), Discrete Wavelet transform (DWT), and Error Correcting Code (ECC). By applying this combination the perceptual quality of the designed scheme is improved but at the same time the computational complexity of the system is high. From the previous discussions on various existing techniques, few research gaps have been identified which are discussed through the following paragraphs: • The EPR watermark needs to be inserted in order to maintain security during transmission to the medical image. It is of utmost importance that the watermark should not deform the medical image of interest. Hence to be on the safe side, in the present work a judicious decision has been taken to use DCT for the embedding process. As its high degree of spectral compaction the DCT property is popular for multidimensional applications; like better perceptual invisibility, adequate robustness, reasonable complexity, and hence less execution time. At the same
Hiding Patient Information in Medical Images …
249
time DCT domain watermarking is popular for hardware implementation because there are a number of fast algorithms exist (Mohamed et al. 2011). • As reported by the previous researchers that the data rates are limited to a maximum value for a given embedding distortion if the wide spectrum of the cover data in EPR hiding is applied. Hence there is a scope in improvement using M-ary modulation over binary signaling principle. The M-ary signaling scheme offers an excellent anti-jamming performance and interference rejection properties. Hence the areas of M-ary signaling principle in Spread Spectrum (SS) approach in DCT remain as one of the subject matters. • In order to detect the watermark properly and to avoid false detection, Error Control Coding (ECC) can be effectively used. Only few of the researchers used ECC for the detection purpose.
1.1 Contribution and Outline of the Paper We proposed a combined approach for embedding EPR using ECC and M-ary modulation in the middle DCT frequency band of the cover image. Using the combined approach of EPR hiding in the image, not only there is a reduction in the bandwidth requirement but also improvement in the system efficacy. (a) Security of the image watermark is enhanced by using ECC and M-ary modulation. By using this combination of techniques the system becomes complex but at the cost of robust improvement. (b) The proposed watermarking scheme performance is evaluated on publically available DICOM dataset. (c) The proposed approach imperceptibility and robustness performances are evaluated using different attacks and at different embedding factor. The rest of the article is comprised as. Section 2 represents the Materials and methods. Results and discussion were presented in Sect. 3. The conclusion drawn from this study is discussed in Sect. 4.
2 Materials and Methods The DICOM brain image database source, the proposed watermarking embedding, and the extraction flow diagram is discussed and enumerated in Fig. 1 and Fig. 2, respectively.
250
R. Agrawal et al.
EPR watermark
Medical cover image as input
Binary sequence
Compute block wise DCT (8x8)
Apply ECC and M-ary modualtion to binary sequence
Embed watermark bit=1
Consider two coefficients B(5,2) and B(4,3) from each block
Yes
B(5,2)=B(5,2) +k/2 B(4,3)=B(4,3)K/2
No
Yes
Swap the coefficients [ B(5,2) B(4,3)]
If B(5,2) >B(4,3)
Yes
If | B(5,2)B(4,3) >’K’
No change
No No No
If B(5,2) < B(4,3)
Yes
Swap the coefficients [ B(5,2) B(4,3)]
Yes
If | B(5,2)B(4,3) >’K’
No
n=n+1
No
All watermark bits embedded
Yes Inverse DCT is applied block wise
Watermarked image
Fig. 1 Flow diagram of watermark embedding
B(5,2)=B(5,2)k/2 B(4,3)=B(4,3) +K/2
No change
Hiding Patient Information in Medical Images …
251
Input watermarked image
Compute Blockwise DCT(8x8)
Choose the same coefficient B(5,2) and B(4,3)
Yes Watermark bit =0
If B(5,2)>B(4,3) No
n=n+1
No
Watermark bit =1
Extract all watermark bits Yes Apply M-ary demodulation
Apply Viterbi decoder
Retrieved EPR watermark
Fig. 2 Flow diagram of extraction algorithm
2.1 Brain Image Dataset The dataset considered in the experiment was obtained from DICOM brain MR image, a standard online database (Chan and Siu 1992). For the purpose of the watermark embedding and extraction, 22 brain tumor infected images were considered in the analysis. However, these datasets do not have any ground truth images.
2.2 Proposed Watermarking Scheme [Embedding and Extraction] This section explains the watermark embedding and extraction process for the proposed watermarking scheme. The watermark used was the Electronic Patient
252
R. Agrawal et al.
Record (EPR) in the image format. This watermark contained all the information regarding the patient such as personal detail of the patient, about the disease the patient is suffering from, detail of the doctor diagnosing the disease and whom doctor to refer, etc. In the watermarking scheme, a watermark was embedded in the whole/total area of the medical image. Discrete Cosine Transformation (DCT) domain was considered as a cover image for embedding and extraction of the watermark in the watermarking schemes. A hybrid combination of Error Control Coding and M-ary modulation was applied to the watermark, after that the coded message was embedded in the mid-band DCT coefficient of the cover medical image. The choice of using ECC is to ensure that the medical image carrying the patient’s information should not get corrupted during transmission. The safe recovery of patient information is important in this situation. So, to recover the maximum amount of text information in a noisy environment, the patient information is coded with ECC techniques. The key idea to use M-ary modulation is to increase the number of possible waveforms being transmitted, but to send only a small subset of the waveforms with more energy transmitted in each waveform. This can be done by grouping log2M bits of the original message and mapping to one of M symbols. By breaking down information to the lowest possible entity, the error probability is minimized since more locations per symbol can be used and simultaneously more locations indicate the improved scope of selecting a higher modulation index. The higher the modulation index indicates improved robustness in terms of transmission reliability through noisy channels. The choice of considering the mid-frequency Cosine transformation band was to provide good imperceptibility and robust to the inserted watermark, i.e., the EPR. Two middle-frequency coefficients locations were chosen arbitrarily for watermark embedding and extraction and the locations were B (5,2) and B (4,3); as shown in Table 1 with grey colored. Watermark bit “1” was embedded if coefficient B (5,2) < B (4, 3); otherwise bit “0” was embedded. The relative modification in the preferred coefficient magnitudes was carried out using the watermark embedding factor “K.” Table 1 DCT block showing 64 coefficients B(1,1)
B(1,2)
B(1,3)
B(1,4)
B(1,5)
B(1,6)
B(1,7)
B(1,8)
B(2,1)
B(2,2)
B(2,3)
B(2,4)
B(2,5)
B(2,6)
B(2,7)
B(2,8)
B(3,1)
B(3,2)
B(3,3)
B(3,4)
B(3,5)
B(3,6)
B(3,7)
B(3,8)
B(4,1)
B(4,2)
B(4,3)
B(4,4)
B(4,5)
B(4,6)
B(4,7)
B(4,8)
B(5,1)
B(5,2)
B(5,3)
B(5,4)
B(5,5)
B(5,6)
B(5,7)
B(5,8)
B(6,1)
B(6,2)
B(6,3)
B(6,4)
B(6,5)
B(6,6)
B(6,7)
B(6,8)
B(7,1)
B(7,2)
B(7,3)
B(7,4)
B(7,5)
B(7,6)
B(7,7)
B(7,8)
B(8,1)
B(8,2)
B(8,3)
B(8,4)
B(8,5)
B(8,6)
B(8,7)
B(8,8)
Hiding Patient Information in Medical Images …
253
3 Results and Discussion In this scheme DICOM brain images have been considered. The DICOM images have an extension as DCM and to be converted into JPEG format. All the medical images used in the study are of size 256 × 256 pixels as host/cover image. A binary image size 32 × 32 pixels is considered as the watermark to be embedded instead of using the actual EPR. In this watermarking scheme, the watermark is embedded in the total medical image. All the experiments were conducted on the processor configuration as 2 GHz, 4 GB RAM using the MATLAB R2013 software platform. The DICOM dataset with the original cover medical images (top) along with their watermarked images (bottom) are shown in Fig. 3. The binary watermark in an image form is shown in Fig. 4. From Fig. 3, it is observed that there is no visual difference between the original image and the watermarked image.
3.1 Performance Measures The robust performance and perceptual/visual quality measurement of the proposed scheme are evaluated using Normalized Cross-Correlation (NCC) (DICOM 2017)
(a)
(b)
(c)
(d)
(a1)
(b1)
(c1)
(d1)
Fig. 3 Original medical image database (top (a–d)) together with their watermarked version (bottom (a1–d1))
Fig. 4 Original binary watermark image
254
R. Agrawal et al.
and Peak Signal to Noise Ratio (PSNR) (DICOM 2017), respectively. Perceptual quality performance metric (PSNR) is evaluated between the watermarked image and the original host image, where robust performance measure (NCC) is evaluated between the retrieved and the original watermark. The equations used for evaluating the performance measures are: a. Peak signal to noise ratio (PSNR): PSNR = 10 log10
Ii,2 j
max MSE
(1)
where (Ii, j )max is the maximum intensity of the original image I(i, j). MSE is mean square error. PSNR is higher for better transformed image. They measure the resemblance between the original and watermarked image. The PSNR value greater than 28 dB is acceptable (Cox and Miller 2002). b. Normalized Cross-Correlation (NCC): It is the measure of similarity between the original watermark image and the retrieved watermark image. The value lies between “0” and “1.” Value “0” indicates no similarity between the two images and value “1” indicates a complete resemblance. It is given by: M N NCC =
j=1 (I (i, j)x F(i, j)) M N 2 i=1 j=1 (I (i, j))
i=1
(2)
3.2 Visual Quality Evaluation The visual quality of the watermark embedding scheme has been studied on the basis of qualitative and quantitative analysis. The qualitative analysis of the scheme is evaluated by observing the DICOM images of Fig. 3. It is observed from Fig. 3 that the original cover DICOM images along with their watermarked images are almost similar. Moreover, the visual quality of the scheme is also evaluated using PSNR. The PSNR performances of the watermarking scheme without signal processing attacks are presented in Table 2. From Table, it is observed that the PSNR values decrease as the value of embedding factor K increases. Furthermore, the average PSNR values of the considered images are observed as 57.45 dB and 54.68 dB at embedding factor K = 10 and 20, respectively. As the PSNR value is above 28 dB (Kallel et al. 2006). Hence, it is proved that the watermarking scheme is capable of producing good quality watermarked images on the basis of qualitative and quantitative analysis.
Hiding Patient Information in Medical Images … Table 2 PSNR verses different embedding factor for different DICOM image
255
Embedding factor K = 10
Images
Embedding factor K = 20
PSNR (dB) Image = 1
59.05
56.28
Image = 2
58.14
55.60
Image = 3
56.08
54.26
Image = 4
56.56
52.59
Average
57.45
54.68
3.3 Robustness Analysis of DICOM Images The robust performance of the watermarking scheme is evaluated in terms of NCC for second datasets (DICOM) against different embedding factor (K = 10 and K = 20). The watermarking scheme robustness validation has been studied by applying different common processing attacks to the watermark image. The various attacks used in the study are: salt and pepper noise (density (D) = 0.01), Gaussian noise (mean (M) = 0 and variance (Var) = 0.01), speckle noise (variance = 0.01), Gaussian Low Pass filter, average filter [3 × 3], median filter [3 × 3], rotation (35 degree), resize (256-128), JPEG (QF = 50), and JPEG 2000 compression attacks. The obtained results in terms of NCC are discussed below in detail: Robustness Analysis against Different Noise Attacks. Various noise attacks were subjected to the watermarked image. The noise considered in the study are the salt and pepper (D = 0.01), Gaussian (M = 0, Var = 0.01) and speckle (Var = 0.01). The NCC values for different noise attacks were enumerated in Table 3 with embedding factor K = 10 and 20 and the corresponding retrieved watermark from Table 3 Robustness analysis of DICOM images against different noise attacks Attacks
NCC values
Average NCC
Image 1 Image 2 Image 3 Image 4 Salt and pepper noise (density = 0.01), 0.9577 (K = 10)
0.9547
0.9597
0.9489
0.9552
Gaussian noise (mean = 0, variance = 0.9554 0.01), (K = 10)
0.9617
0.9532
0.9546
0.9562
Speckle noise (variance = 0.01), (K = 0.8999 10)
0.9278
0.9131
0.9055
0.9115
Salt and pepper noise (density = 0.01), 0.9598 (K = 20)
0.9605
0.9632
0.9582
0.9604
Gaussian noise (mean = 0, variance = 0.9590 0.01), (K = 20)
0.9621
0.9585
0.9600
0.9599
Speckle noise (variance = 0.01), (K = 0.9001 20)
0.9365
0.9055
0.9235
0.9164
256
R. Agrawal et al.
the attacked DICOM images for K = 10 were shown in Fig. 5. The average NCC of the extracted watermark is 0.9552 and 0.9604 when salt and pepper noise was applied to the same dataset at embedding factor (K = 10 and K = 20), respectively.
Fig. 5 1 (a–d) watermarked image against salt and pepper noise. 1 (a1–d1) Extracted watermark against salt and pepper noise. 2 (a–d) Watermarked image against Gaussian noise. 2 (a1–d1) Extracted watermark against Gaussian noise. 3 (a–d) Watermarked image against speckle noise. 3 (a1–d1) Extracted watermark
Hiding Patient Information in Medical Images …
257
Table 4 Robustness analysis of DICOM images against different filtering attack Attacks
NCC values
Average NCC
Image 1
Image 2
Image 3
Image 4
Gaussian low pass (K = 10)
0.9554
0.9669
0.9652
0.9407
0.9570
Averaging [3 × 3] (K = 10)
0.9355
0.9466
0.9576
0.9497
0.9473
Median [3 × 3] (K = 10)
0.9490
0.9527
0.9534
0.9552
0.9525
Gaussian low pass (K = 20)
0.9595
0.9696
0.9652
0.9413
0.9589
Averaging [3 × 3] (K = 20)
0.9412
0.9490
0.9578
0.9518
0.9499
Median [3 × 3] (K = 20)
0.9512
0.9602
0.9610
0.9554
0.9569
Similarly, the average NCC of the extracted watermark is 0.9562 and 0.9599 when Gaussian noise was applied to the watermarked image to the same dataset for the same embedding factor, respectively. Further, the average NCC value is 0.9115 and 0.9164, respectively, when speckle noise is. It is observed from Table 3 that the watermarking scheme is robust against different noise attacks, as in all the cases the average NCC value is higher than 0.7 (Kallel et al. 2006). Robustness Analysis against Different Filtering Attacks. Different filtering attacks have been subjected to the watermarked image. The filter considered are the Gaussian Low Pass, average filter [3 × 3] and median filter [3 × 3]. The average NCC of the extracted watermark is 0.9570 and 0.9589 when Gaussian LPF is applied to the watermarked image at embedding factor K = 10 and 20, respectively. The average NCC of the extracted watermark were obtained as 0.9473 and 0.9499 when average filter [3 × 3] kernel is applied to the watermarked image for K = 10 and 20, respectively. Also, the average NCC value as 0.9525 and 0.9569 when the median filter with [3 × 3] kernel is applied for different values of K, respectively, as already discussed. The NCC values for different filtering attacks has enumerated in Table 4 for different values of K and the retrieved watermark from the same attack at K = 10 for the same set of images are shown in Fig. 6. Robustness Analysis against Different Geometrical Attacks. The watermarked images have been tested for different geometrical attacks, namely, rotation and resizing. The 35 angle rotation and [256-128-256] resizing attacks are applied to the watermarked image, and the results were shown in Table 5 and Fig. 7. The average NCC values of the extracted watermark are obtained as 0.9241 and 0.9255 against rotation attack, respectively, for different values of K (10 and 20), and the average NCC value against resizing attacks for considered values of K was obtained as 0.9301 and 0.9389, respectively, are shown in Table. The extracted watermark for K = 10 was shown in Fig. 7. Table and figure reveal that the watermarking scheme is highly robust against geometrical attacks. Robustness Analysis against Different Compression Attacks. The watermarked images have been tested for different compression attacks, namely, JPEG and JPEG 2000. These attacks were applied to the watermarked image for different K values, and the results were shown in the Table and figure. The average NCC values of the extracted watermark are obtained as 0.9545 and 0.9560 against JPEG
258
R. Agrawal et al.
Fig. 6 1 (a–d) Watermarked image against Gaussian LPF. 1 (a1–d1) Extracted watermark against Gaussian LPF. 2 (a–d) Watermarked image against average [3 × 3]. 2 (a1–d1) Extracted against watermark average [3 × 3]. 3 (a–d) Watermarked image against median [3 × 3]. 3 (a1–d1) Extracted watermark against median [3 × 3]
Hiding Patient Information in Medical Images …
259
Table 5 Robustness analysis of DICOM images against geometrical attack Attacks
NCC values Image 1
Image 2
Average NCC Image 3
Image 4
Rotation (35 degree) (K = 10)
0.9349
0.9318
0.901
0.9278
0.9241
Resizing (256-128-256) (K = 10)
0.9408
0.9089
0.9364
0.9345
0.9301
Rotation (35 degree) (K = 20)
0.9350
0.9328
0.9031
0.9312
0.9255
Resizing (256-128-256) (K = 20)
0.9508
0.9112
0.9404
0.9532
0.9389
Fig. 7 1 (a–d) Watermarked image against rotation (35 degree). 1 (a1–d1) Extracted watermark against rotation (35 degree). 2 (a–d) Watermarked image against resizing [128-256-128]. 2 (a1–d1) Extracted watermark against resizing [128-256-128]
compression attack, respectively, for different values of K, and the average NCC value against JPEG 2000 compression for considering values of K were obtained as 0.9584 and 0.9630, respectively, and are shown in Table 6. The extracted watermark for the same experimental images with K = 10 was shown in Fig. 8.
260
R. Agrawal et al.
Table 6 Robustness analysis of DICOM images against compression attack Attacks
NCC values
Average NCC
Image 1
Image 2
Image 3
Image 4
JPEG (QF = 50) (K = 10)
0.9466
0.9664
0.9586
0.9464
JPEG 2000 (K = 10)
0.9532
0.9645
0.9640
0.9520
0.9584
JPEG (QF = 50) (K = 20)
0.9483
0.9672
0.9601
0.9485
0.9560
JPEG 2000 (K = 20)
0.9542
0.9681
0.9697
0.9602
0.9630
0.9545
Fig. 8 1 (a–d) Watermarked image against JPEG (QF = 50). 1 (a1–d1) Extracted watermark. 2 (a–d) Watermarked image against JPEG 2000 compression. 2 (a1–d1) Extracted watermark
4 Conclusion Now a days, electronic health care system is becoming popular. In this system, the remote area doctors send the patient information interleaving with the medical image to the expert physicians for a second opinion viz open network. To maintain the security to the EPR data medical image watermarking are a potent solution. This paper presents a new watermarking method in the application of medical image where
Hiding Patient Information in Medical Images …
261
the EPR data is embedded in the host image using DCT transform. To enhance the watermarking scheme robustness, initially the binary EPR information is encoded using ECC and then M-ary modulation is applied. The attractive features of the proposed work are summarized as follows: • The watermarking techniques using ECC, M-ary modulation, and DCT exhibit better performance in terms of imperceptibility and robustness. • Security of the image watermark is enhanced by using ECC and M-ary modulation. By using this combination of techniques the system becomes complex but at the cost of robust improvement. Various techniques were combined to make a balance between the imperceptibility and robustness of the watermarking system. Due to hybrid combination the computational complexity of the system is high and needs to be investigated separately in future communication.
References Chan YH, Siu WC (1992) On the realization of discrete cosine transform using the distributed arithmetic. IEEE Trans Circ Syst I: Fundam Theory Appl 39(9):705–712 Cox IJ, Miller ML (2002) The first 50 years of electronic watermarking. EURASIP J Adv Sig Process 2002(2):820936 Das S, Kundu MK (2011) Hybrid contourlet-DCT based robust image watermarking technique applied to medical data management. In: International conference on pattern recognition and machine intelligence, pp 286–292. Springer, Berlin, Heidelberg DICOM Samples Image Sets, http://www.osirix-viewer.com/. Accessed Sept 2017 Hajjaji MA, Mtibaa A, Bourennane E (2011) A watermarking of medical image: method based “LSB”. J Emerg Trends Comput Inform Sci 2(12) Kallel IF, Kallel M, Bouhlel MS (2006) A secure fragile watermarking algorithm for medical image authentication in the DCTdomain. In: Proceedings of IEEE conference on information and communication technology, ICTTA’06 Syria, pp 2024–2029 Kannammal A, Subha Rani S (2014) Two level security for medical images using watermarking/encryption algorithms. Int J Imaging Syst Technol 24(1):111–120 Kutter M (1999) Performance improvement of spread spectrum based image watermarking schemes through M-ary modulation. Lecture notes in computer science, Brisbane, Germany Mehto A, Mehra N (2016) Adaptive lossless medical image watermarking algorithm based on DCT & DWT. Procedia Comput Sci 78:88–94 Mostafa SA, El-Sheimy N, Tolba AS, Abdelkader FM, Elhindy HM (2010) Wavelet packets-based blind watermarking for medical image management. Open Biomed Eng J 4:93 Nambakhsh MS, Ahmadian A, Zaidi H (2011) A contextual based double watermarking of PET images by patient ID and ECG signal. Comput Methods Programs Biomed 104(3):418–425 Singh AK, Kumar B, Dave M, Mohan A (2015a) Robust and imperceptible dual watermarking for telemedicine applications. Wireless Pers Commun 80(4):1415–1433 Singh AK, Kumar B, Dave M, Mohan A (2015b) Multiple watermarking on medical images using selective discrete wavelet transform coefficients. J Med Imaging Health Inform 5(3):607–614 Swaraja K (2017) A hybrid secure watermarking technique in telemedicine. Int J Eng Technol 9(3):265–270 Zear A, Singh AK, Kumar P (2018) A proposed secure multiple watermarking technique based on DWT, DCT and SVD for application in medicine. Multimedia Tool Appl 77(4):4863–4882
Segmented Lung Boundary Correction in Chest Radiograph Using Context-Aware Adaptive Scan Algorithm Tej Bahadur Chandra, Kesari Verma, Deepak Jain, and Satyabhuwan Singh Netam
Abstract Chest X-ray (CXR) is the most popular imaging modality used for preliminary diagnosis of pulmonary diseases. In automatic computer-aided diagnosis (CAD), the number of false-positive cases can be reduced by segmenting out the normal anatomical structures. Demarcating lung parenchyma on CXR image is challenging due to complex anatomical structures of the human thoracic cavity. The extracted lungs boundary suffers from undesirable artifacts such as ridges and pits. This paper presents an algorithm to adaptively scan the inner lung boundary to correct the undesirable artifacts. Further, the algorithm is context-aware, it takes care of normal cavity due to the aortic knuckle and diaphragm borders. The algorithm is tested on 138 binary lung mask extracted from digital CXR images from Montgomery dataset. The quantitative, qualitative, statistical results reveal that the proposed algorithm outperforms the existing method. The average increase in segmentation accuracy is 2.561%. Keywords Chest X-ray · Lung segmentation · Radiography · Anatomical atlas · Thoracic disease · Costophrenic angle
T. B. Chandra (B) · K. Verma Department of Computer Applications, National Institute of Technology, Raipur, India e-mail: [email protected] K. Verma e-mail: [email protected] D. Jain · S. S. Netam Department of Radiodiagnosis, Pt. Jawahar Lal Nehru Memorial Medical College, Raipur, India e-mail: [email protected] S. S. Netam e-mail: [email protected] © Springer Nature Singapore Pte Ltd. 2021 A. A. Rizvanov et al. (eds.), Advances in Biomedical Engineering and Technology, Lecture Notes in Bioengineering, https://doi.org/10.1007/978-981-15-6329-4_23
263
264
T. B. Chandra et al.
1 Introduction Chest X-ray (CXR) imaging modality is an important, noninvasive clinical adjunct for preliminary diagnosis of interstitial lung diseases. Automatic computer-aided diagnosis (CAD)of ambiguous thoracic disease manifestations on radiographic images is challenging. The elusive normal anatomical structures and vague abnormal opacities in lung fields require localized deep feature analysis to characterize normal and abnormal CXR images. In this context, accurate identification of lung parenchyma regions provides rich structural information like size, shape, lung volume, etc., for subsequent phases of automated diagnosis (Candemir et al. 2014; Jaeger et al. 2014). The accurate delineation of lung boundary is challenging due to an overlapping complex anatomical structure like rib cage, heart apex, hilum, aortic knuckle, and strong edges of clavicle bones (Chandra and Verma 2020; Chondro et al. 2018) as shown in Fig. 1a, b. Several state of art methods have been proposed over past few years for lung parenchyma segmentation and automated analysis of CXR images. Ginneken et al. (2001) performed a survey on 150 papers and suggested that CAD-based analysis on segmented lungs significantly reduce the false positives and increase the effectiveness of abnormality detection. Shi et al. (2006) used modified SIFT method and patient-specific shape statistics to segment lung fields from a series of chest radiograph of the same patient and achieved segmentation accuracy of 93.8% using the proposed method which is significantly higher than the performance obtained in ASM method (92.6%) (Cootes and Taylor 2004) and Snake method (78.7%) (Kass et al. 1988). Annangi et al. (2010) utilized canny edges features, CP angle corner feature to extract the lung air cavity regions from
Fig. 1 a Normal chest X-ray image with marked lung boundary (1-clavicle bone, 2-left and right hilum, 3-gastric bubble). b Corresponding normal binary lung mask (1-aortic knuckle, 2-heart apex, 3-left and right diaphragm border, 4-left and right costophrenic angle). c–h Binary lung mask showing undesirable ridges and pits due to strong bone edges or consolidations near inner lung boundary
Segmented Lung Boundary Correction in Chest Radiograph …
265
posterior-anterior (PA) CXR images and attains dice similarity coefficient value of 0.88 ± 0.07. Jaeger et. al. (2012) used collective lung shape modal, intensity modal, and segmentation mask to perform accurate segmentation. The proposed method provides AUC of 83.12%. Candemir et al. (2012) proposed the lung modalbased graph cut method for lung boundary detection for segmenting lung air cavity and observed 91% of segmentation accuracy on JSRT dataset, further the enhanced version using SIFT flow-based nonrigid deformable registration method to automatically compute the patient-specific binary lung mask is presented in Candemir et al. (2014) attaining higher segmentation accuracy of 95.4% (JSRT dataset), 94.1% (Montgomery dataset), and 91.7% (Maryland dataset). A recent method for lung segmentation using ultrametric contour map and structured edge detector is proposed in Yang et al. (2018) achieving a mean Jaccard index of 95.2%. Segmented lung regions play an important role in the quantitative estimation of radiographic indices like lung volumes, varying heart dimension, irregular lung shape, cardiothoracic (CT) ratio, costophrenic angle, and other pathologies which can be used as a measure to detect cardiomegaly, pneumothorax, atelectasis, pleural effusion, and other chronic abnormalities. The automatic localization of costophrenic recesses and measurement of costophrenic angle on CXR images are described in Maduskar et al. (2013), Armato et al. (2004), Armato et al. (1998), Hasan et al. (2012), Dallal et al. (2017). A very limited number of methods have been reported in literatures related to lung boundary correction using computed tomography (CT) images. Yim and Hong (2008) proposed curvature-based lung boundary correction method in CT images using 3D volume of interest (VOI) refinement and achieved overlapping area ratio is 96%. Pu et al. (2008) use geometric smoothening of lungs border to include juxtapleural nodules in CT images while reducing the over-segmentation. Liang et al. (2017) proposed boundary tracing algorithm which detects the inflection point more accurately than others. The proposed method shows the 6.46% increase in segmentation precision compared to existing state of art methods. Singadkar et al. (2018) use dominant point marching algorithm to connect concave and convex regions. The obtained average volumetric overlap is 96.97%. Despite of several improvements in lung segmentation algorithms, boundary correction in CXR images remains an unexplored area. Artifacts like ridges and pits degrade the segmentation accuracy subsequently affecting the performance of CAD-based automated systems. The undesirable artifacts are due to strong edges of the bones or due to disease opacification near lung boundary regions. In this study, a new context-aware adaptive scan correcting algorithm is proposed to adaptively scan the inner lung boundary in patient-specific binary lung mask of CXR image to correct the undesirable ridge and pit as shown in Fig. 1c–h. The working of the algorithm is derived from the core idea presented in paper (Sezaki and Ukena 1973). The algorithm is context-aware, it takes care of the lower diaphragm curve and normal pits due to the aortic knuckle near the upper inner border of the right lung. The rest of the paper is structured as follows. Section 2, elaborates the detailed implementation of the algorithm. In Sect. 3, the brief description of the materials and method used in this study is presented. The experimental setup, quantitative and qualitative evaluation of results are described in Sect. 4. In Sect. 5, the detailed discussion is presented followed by a conclusion and future direction in Sect. 6.
266
T. B. Chandra et al.
2 Proposed Algorithm The proposed context-aware adaptive scan algorithm is an enhancement of method presented in paper (Sezaki and Ukena 1973). Moreover, this proposed algorithm is used to enhance the segmentation accuracy of binary lung mask-based segmentation techniques. In this study, the algorithm uses binary lung mask obtained from the SIFT flow-based nonrigid registration-driven lung segmentation method (Candemir et al. 2014) for correcting ridges and pits on the inner boundary of segmented lung mask.The working principle of the proposed method is illustrated in Fig. 2a, b and a detailed description is presented in Algorithm 1. The algorithm initially set the scan lines at given interval si from top to bottom scan limit (B pt ) of binary lung mask (Fig. 2a). Subsequently for each scan line, left and right inner lung border point pi is retrieved. However, in case of left lung, the scanning starts after skipping the topskip percent area (usually greater than 30%) from the top to accommodate the normal cavity structure due to the aortic knuckle as shown in Fig. 2a. The algorithm adaptively scan from the point pi by placing two more points pi+1 , and pi+2 on the scan line at interval si and 2si, respectively, as shown in Fig. 2b. If the point pi+1 lies outside the range of ( pi ± d1) and point pi+2 lies within the range of ( pi ± d2) then the point pi+1 is considered either pit or ridge and new border point is computed by averaging the first and last value as described in line 14 of the Algorithm 1. Moreover, if the point pi+1 falls in the range of ( pi ± d1) but the point pi+2 lies outside the range of ( pi ± d2) then the point pi+1 is considered as normal border point and the algorithm will simply skip the point. Further, if the points pi+1 and pi+2 both lies outside the range of ( pi ±d1) and ( pi ±d2), respectively, then the algorithm follows the recursive approach with adjusting the learning parameters μ2 + γ and adapted scan interval si + ω to deal with large pits or ridges as described in line 16, 17, and 19 of Algorithm 1. Finally, the algorithm stops when the scan interval
Fig. 2 a The binary lung mask showing a scan line on the area of interest. b Working principle of the algorithm to correct the ridges and pits in lungs boundary using binary lung mask
Segmented Lung Boundary Correction in Chest Radiograph …
267
si > si m . The values of d1 and d2 are adjusted automatically based on the scan interval and the learning parameters μ1 andμ2 as described in line 8 and 9 of the Algorithm 1. The values of learning parameters are determined empirically after a series of repeated experimentation.
268
T. B. Chandra et al.
3 Materials and Methods 3.1 Dataset In this study, the Montgomery Set (Candemir et al. 2014; Jaeger et al. 2014) is used to evaluate the performance of the proposed method. The dataset contains a total 138 digital CXR images comprising 58 images with tuberculosis manifestation and 80 images with normal findings. All the X-ray images are in portable network graphics (PNG) file format of dimension 4020 × 4892 or 4892 × 4020 with a bit-depth of 12 bits. The pixels are linearly spaced in both horizontal and vertical direction with spacing 0.0875 mm. The dataset is created and maintained by National Library of Medicine in collaboration with the Department of Health and Human Services, Montgomery County, Maryland, USA.
3.2 Lung Segmentation The patient-specific binary lung mask in this study is obtained from nonrigid registration-driven robust lung segmentation method proposed in Candemir et al. (2014). The detailed working of the method is shown in Fig. 3. The method is based on content-based image retrieval (CBIR) technique to get the patient-specific adaptive lung mask. It uses Bhattacharyya shape similarity index to get the most similar reference images to patient X-ray, at the same time partial radon transform is used to handle the little affine variation in input X-rays images. The corresponding lung mask of top n-ranked similar reference images are registered using deformable SIFT
Fig. 3 Three stages of nonrigid registration-driven lung segmentation method
Segmented Lung Boundary Correction in Chest Radiograph …
269
flow registration method to get the patient-specific average lung mask. Finally, the graph cut method is used to cut the original CXR image to get the segmented lungs.
3.3 Evaluation Measures To evaluate and compare the performance of the proposed algorithm qualitative and quantitative evaluation measures are used as described in the following subsections. Quantitative Evaluation. The Jaccard similarity coefficient (Panigrahi et al. 2019) as described in Eq. 1 is used to measure the overlap between the corrected binary lung mask (Scorrected ) and the corresponding gold standard ground truth (SGT ) mask over all the pixels. Where true positive (TP) represents the correctly identified foreground pixel, false positive (FP) represents the pixels misclassified as foreground and false negative (FN) represent the pixels misclassified as background pixels. JaccardIndex(JI) =
|TP| |Scorrected ∩ SGT | = |FP| + |TP| + |FN| |Scorrected ∪ SGT |
JaccardDistacne(JD) = 1 −
|Scorrected ∩ SGT | |Scorrected ∪ SGT |
(1) (2)
Similarly, dice’s index (DI) (Dice 1945) as described in Eq. 3 is also computed to measure the area of overlap. Dices Index(DI) =
2|TP| |Scorrected ∩ SGT | = |Scorrected | + |SGT | 2|TP| + |FN| + |FP|
(3)
Qualitative Evaluation. The visual assessment of the corrected binary lung mask is important since the topography of normal anatomical structure plays an important role in abnormality detection. The lung mask before and after correction is fused (overlapped) with white regions representing the lung mask before correction and pink region representing the lung mask after correction as shown in Fig. 7. These fused lung mask images along with the original CXR images ware provided to an experienced radiologist for visual assessment. The opinion of the radiologist regarding the accuracy of ridges or pits correction using the proposed algorithm is recorded as the level of satisfaction (in %). Statistical Analysis. Statistical analysis plays an important role in medical image analysis to examine the significance of a method based on the several experimental runs. In this study, Z-test statistics for two samples are used to assess the statistical significance of the proposed algorithm over the existing method (Agnihotri et al. 2017). The Z-score as described in Eq. 4 uses mean and variance of the two populations to accept or reject the null hypothesis (H0 ) based on the p-value and critical value at specified significance level (α).
270
T. B. Chandra et al.
Z=
M1 − M2 − 2 σ1 σ2 + n22 n1
(4)
where, M1 and M2 are the means of Jaccard index of existing and proposed method, respectively, is the hypothesized difference between the population means, σ1 and σ2 are the standard deviation of two populations and n 1 and n 2 are the sizes of two samples. Where the null hypothesis (H0 ) denotes the mean segmentation performance of proposed algorithm and the existing method is equal and the alternate hypothesis denotes the performance of proposed algorithm is significantly higher than the existing method. The assumed null and alternate hypotheses are described in Eq. 5 and Eq. 6, respectively. H0 : M1 = M2
(5)
H1 : M1 < M2
(6)
4 Experimental Setup This section presents the topological design of an experimental setup to test and compare the performance-proposed algorithm as shown in Fig. 4. The algorithm is tested on 138 binary lung masks extracted using nonrigid registration-based lung segmentation method from Montgomery dataset. Further, the twofold evaluation measures (quantitative and qualitative) are used to evaluate the performance.
Fig. 4 Topological design of experimental setup
Segmented Lung Boundary Correction in Chest Radiograph …
271
5 Results and Discussion This section presents the results of the quantitative evaluation and visual assessment of the proposed algorithms. The response of the proposed method while dealing with ridges and pits are illustrated in Fig. 5. It can be observed from Fig. 5a–h that the point pi+1 lies outside the range of pi ±d1 while pi+2 lies within the range of pi ±d2 showing the presence of either ridge or pits.The computed new boundary point is the averaged value of two points pi and pi+2 shown as red-colored cross mark in each image patch. Furthermore, if the size of artifact is large that remains undetected by scan lines at smaller intervals, the algorithm gradually increases the scan interval by ω and widens the range of d2 by γ in each iteration as observed from Fig. 5d. The results of quantitative evaluation measures: average Jaccard index (JI), Jaccard distance (JD), and dice’s index (DI) to measure the performance of the algorithm are shown in Table 1. It can be observed from Table 1 that average Jaccard index (overlap measure) between segmented binary lung mask (before correction) and gold standard ground truth (GT) is 0.8960. However, after correcting the lung
Fig. 5 Scan correcting the ridges and pits of binary lung mask using adaptive scan lines
Table 1 Segmentation performance measure using average Jaccard index (JI), average Jaccard distance (JD), and average dice’s index (DI) of binary mask before and after lung mask correction Evaluation measures
Before correction
After correction
Average Jaccard index
0.8960
0.9216
Average Jaccard distance
0.1040
0.0784
Average Dice’s index
0.9451
0.9592
272
T. B. Chandra et al.
mask the Jaccard index is increased by 0.0256. The improvement in the performance is due to the fact that the ridges and pits due to strong edges of the normal anatomical structure are corrected by the proposed algorithm as shown in Fig. 8. The increase in average dice’s index and decreased average Jaccard distance after lung mask correction confirms the promising performance of the proposed algorithm. Figure 6 shows the plot of the Jaccard similarity index before and after correcting the individual binary lung mask. From the plotted graph it can be observed that the overlap area between corrected lung mask and gold standard ground truth mask is significantly better than that of before correction for each individual lung mask. The performance comparison of the newly proposed and existing method is performed using Receiver Operating Characteristic (ROC) curve as shown in Fig. 7b. From the area under ROC curve analysis it can be observed that the proposed boundary correction method has improved the lung segmentation accuracy attaining AUC of 0.908.
Fig. 6 The plot of the Jaccard similarity index (before and after correction) between individual binary lung mask and gold standard ground truth
Fig. 7 a Correlation coefficient plot between observed quantitative result and result of subjective evaluation by radiologist. b The receiver operating characteristic (ROC) curve plot for proposed algorithm and existing method
Segmented Lung Boundary Correction in Chest Radiograph …
273
The Pearson’s correlation coefficient (r) plot shown in Fig. 7a shows the strength of linear relationship between observed quantitative results and the results of the subjective evaluation performed by the radiologist. The value of r ranges between −1 to 0 to +1 for negatively correlated, not correlated and, positively correlated, respectively. The obtained value of correlation coefficient r = 0.9447 is very close to +1 which denotes that the result of subjective evaluation performed by the radiologist is positively correlated with the results obtained by quantitative evaluation. The qualitative evaluation of corrected binary lung mask (as shown in Fig. 8) is performed by an expert radiologist by rating (in %) based on visual appearance of the corrected lung mask (as shown in Fig. 8). As per the radiologist suggestion the segmented part of lungs using existing method (anatomical atlas-based segmentation method Candemir et al. 2014) specially in regions with strong bone or disease responses are not properly segmented that affects the computation of features for diseases diagnosis. The proposed method resolves this issue and improve the result based on visual assessment attaining average visual score by 92.53%.The average improvement in visual segmentation result is 2.93% compared to existing method. Moreover, as suggested by the radiologist the algorithm is lacking in some cases, especially when the ridges or pits are large enough that remains undetected or partially detected by the algorithm as shown in Fig. 8b, c, e, h, i, l which can be the future research direction. The statistical analysis performed using Z-test method for two sample means on obtained quantitative result at 0.05 level of significance (α) is shown in Table 2. From the Table 2, it can be observed that the obtained p-value (0.00006) is significantly smaller than the given significance level (α = 0.05) which clearly rejects the null hypothesis. Furthermore the value of calculated Z-score is less than the Z-score at critical point which confirms that acceptance of alternate hypothesis that verifies the statistical significance of the proposed algorithm over the existing method.
Fig. 8 Corrected binary mask images using the proposed algorithm (white regions show the actual segmentation having ridges and pits while the pink region shows the corrected binary mask)
274
T. B. Chandra et al.
Table 2 Z-test scores (left tail test) for two sample means at 0.05 level of significance (α) for 138 observations in each samples Statistics value Z-score (calculated)
−3.83981
Z-score (critical)
1.64485
p-value
0.00006
6 Conclusion In this paper, we have proposed the context-aware adaptive scan algorithm for correcting the artifacts (ridges and pits) in binary lung masks. The proposed method uses adaptive scanning for eliminating the artifacts in the inner lung border improving the segmentation accuracy. The quantitative evaluation measures reveal the better performance of the proposed algorithm attaining an average 2.561% increase in segmentation accuracy compared to existing methods without lung mask correction. Moreover, the results of visual evaluation performed by the radiologist confirm that the corrected binary mask more accurately represents the lung fields. The statistical analysis performed using Z-test score for two sample means confirms the statistical significance of the proposed algorithm. In future, the proposed algorithm can be enhanced in several ways: (1) Costophrenic recess and diaphragm contours can be corrected intelligently. (2) Artifacts near normal aortic knuckle cavity can be improved. (3) The algorithm can be evaluated on larger datasets. (4) Intra- and inter-operator and observer analysis can also be performed.
References Agnihotri D, Verma K, Tripathi P (2017) Variable global feature selection scheme for automatic classification of text documents. Expert Syst Appl 81:268–281 Annangi P, Thiruvenkadam S, Raja A, Xu H, Sun XSX, Mao LML (2010) A region based active contour method for x-ray lung segmentation using prior shape and low level features. In: 2010 IEEE International Symposium on Biomedical Imaging: From Nano to Macro, pp 892–895 Armato III SG, Giger ML, MacMahon H (1998) Computerized delineation and analysis of costophrenic angles in digital chest radiographs. Acad Radiol 5(5):329–335. https://doi.org/10. 1016/S1076-6332(98)80151-7 Armato III SG, Giger ML, MacMahon H (2004) Method and system for the automated delineation of lung regions and costophrenic angles in chest radiographs Candemir S, Jaeger S, Palaniappan K, Antani S, Thoma G (2012) Graph cut based automatic lung boundary detection in chest radiographs. In: IEEE healthcare technology cconference: translational engineering in health and medicine, pp 31–34 Candemir S, Jaeger S, Palaniappan K, Musco JP, Singh RK, Xue Z, Karargyris A, Antani S, Thoma G, McDonald CJ (2014) Lung segmentation in chest radiographs using anatomical atlases with nonrigid registration. IEEE Trans Med Imaging 33:577–590
Segmented Lung Boundary Correction in Chest Radiograph …
275
Chandra TB, Verma K (2020) Pneumonia detection on chest X-Ray using machine learning paradigm. In: Proceedings of 3rd international conference on computer vision and image processing. Springer, Singapore, pp 21–33. https://doi.org/10.1007/978-981-32-9088-4_3 Chondro P, Yao CY, Ruan SJ, Chien LC (2018) Low order adaptive region growing for lung segmentation on plain chest radiographs. Neurocomputing. 275:1002–1011 Cootes TF, Taylor CJ (2004) Statistical models of appearance for computer vision. Available on: https://www.face-rec.org/Algorithms/AAM/app_models.pdf. Accessed 24 July 2020 Dallal AH, Agarwal C, Arbabshirani MR, Patel A, Moore G (2017) Automatic estimation of heart boundaries and cardiothoracic ratio from chest x-ray images. In: Medical imaging 2017: computer-aided diagnosis, p 101340K Dice LR (1945) Measures of the amount of ecologic association between species. Ecology 26:297– 302 Hasan MA, Lee S-L, Kim D-H, Lim M-K (2012) Automatic evaluation of cardiac hypertrophy using cardiothoracic area ratio in chest radiograph images. Comput Methods Programs Biomed 105:95–108 Jaeger S, Karargyris A, Antani S, Thoma G (2012) Detecting tuberculosis in radiographs using combined lung masks. In: 2012 Annual international conference of the IEEE on engineering in medicine and biology society (EMBC), pp 4978–4981 Jaeger S, Karargyris A, Candemir S, Folio L, Siegelman J, Callaghan FM, Xue Z, Palaniappan K, Singh RK, Antani SK et al (2014) Automatic tuberculosis screening using chest radiographs. IEEE Trans Med Imaging 33:233–245 Kass M, Witkin A, Terzopoulos D (1988) Snakes: active contour models. Int J Comput Vis 1:321– 331 Liang J, Jiang H, Ma L, Liu Y, Toshiya N (2017) A new boundary correction method for lung parenchyma. In: Eighth international conference on graphic and image processing (ICGIP 2016), vol 10225. International Society for Optics and Photonics, p 1022529. https://doi.org/10.1117/ 12.2266086 Maduskar P, Hogeweg L, Philipsen R, van Ginneken B (2013) Automated localization of costophrenic recesses and costophrenic angle measurement on frontal chest radiographs. In: Medical imaging 2013: computer-aided diagnosis, vol 8670. International Society for Optics and Photonics, p 867038. https://doi.org/10.1117/12.2008239 Panigrahi L, Verma K, Singh BK (2019) Ultrasound image segmentation using a novel multiscale Gaussian kernel fuzzy clustering and multi-scale vector field convolution. Expert Syst Appl 115:486–498 Pu J, Roos J, Yi CA, Napel S, Rubin GD, Paik DS (2008) Adaptive border marching algorithm: Automatic lung segmentation on chest CT images. Comput Med Imaging Graph 32:452–462 Sezaki N, Ukena K (1973) Automatic computation of the cardiothoracic ratio with application to mass screening. IEEE Trans Biomed Eng:248–253 Shi Y, Qi F, Xue Z, Ito K, Matsuo H, Shen D (2006) Segmenting lung fields in serial chest radiographs using both population and patient-specific shape statistics. In: International conference on medical image computing and computer-assisted intervention. Springer, Berlin, Heidelberg, pp 83–91. https://doi.org/10.1007/11866565_11 Singadkar G, Mahajan A, Thakur M, Talbar S (2018) Automatic lung segmentation for the inclusion of juxtapleural nodules and pulmonary vessels using curvature based border correction. J King Saud Univ - Comput Inf Sci Van Ginneken B, Ter Haar Romeny BM, Viergever MA (2001) Computer-aided diagnosis in chest radiography: a survey. IEEE Trans Med Imaging 20:1228–1241 Yang W, Liu Y, Lin L, Yun Z, Lu Z, Feng Q, Chen W (2018) Lung field segmentation in chest radiographs from boundary maps by a structured edge detector. IEEE J Biomed Heal Inform 22:842–851 Yim Y, Hong H (2008) Correction of segmented lung boundary for inclusion of pleural nodules and pulmonary vessels in chest CT images. Comput Biol Med 38:845–857
Effect of Temperature and Titania Doping on Structure of Hydroxyapatite Yash Chopra, Rajesh Kumar, and Howa Begam
Abstract Hydroxyapatite (HAp) doped with titanium dioxide was prepared by wet chemical route. Thus prepared HAp was heat-treated at 800 °C and sintered at 1250 °C. To study the effect of dopant and temperature on crystallographic structure, X-Ray diffraction, and Fourier transform infrared spectroscopy (FTIR) was performed. It was observed that all materials were in single phase and few cases peak shifting were observed. Dopant causes distortion of hexagonal crystal structure, which lessened the lattice parameter and crystallinity. Temperature had also role in changes in unit cell volume and percentage of crystallinity. Similar effect was observed in FTIR result. Keywords Titanium dioxide · Hydroxyapatite · XRD · FTIR · Unit cell volume · Crystallinity
1 Introduction Calcium phosphate-based bioceramics are extensively used as implant biomaterial due to its capacity of bone growth and biocompatibility (Kalita et al. 2007). Synthetic hydroxyapatite has engrossed special consideration as bone and teeth implant material. Beside its advantages, there are so many shortcomings as implant material like its inferior osteoinductivity, brittleness (Bodhak et al. 2011). Enhancement of osteoinductivity is still a significant challenge. Other limitations are including high percentage of crystallinity, that causes nondegradability of HAp when it is implanted into the body (Geng et al. 2016). Therefore, there is an emergent concern to make a new generation biomaterial using HAp, that will reduce the shortcomings of synthetic hydroxyapatite and it will help in faster healing after implantation. Y. Chopra · R. Kumar (B) · H. Begam Department of Biomedical Engineering, National Institute of Technology Raipur, Raipur, India e-mail: [email protected] H. Begam e-mail: [email protected] © Springer Nature Singapore Pte Ltd. 2021 A. A. Rizvanov et al. (eds.), Advances in Biomedical Engineering and Technology, Lecture Notes in Bioengineering, https://doi.org/10.1007/978-981-15-6329-4_24
277
278
Y. Chopra et al.
The use of metallic ion as substitute element is one approach for betterment of synthetic HAp in terms of mechanical properties, crystallinity, etc. Geng et al. (2016) showed that magnesium (Mg2+ ) and strontium (Sr2+ ) ions (Mg/Sr/(Ca + Mg + Sr) concentration: 10/20 mol%) influenced the mineral metabolism at the time of bone remodeling process and its also enhanced osteoblast cell proliferation. Zinc dopant improved new bone formation and it showed more new bone formation compared to undoped HAp when it was implanted in rabbit model (Bhattacharjee et al. 2014). Kalita et al. used zinc and magnesium dopant during the sol–gel synthesis of nano hydroxyapatite and they showed that the zinc and magnesium dopant improved the surface hardness and compression strength of nano-HAp (Kalita and Bhatt 2007). Stani´c et al. (2010) used copper and zinc ions as dopant and concluded on the basis of their antimicrobial test results that all metal doped HAp exhibit reduction of viable cell of Escherichia coli, Staphylococcus aureus, and Candida albicans. Ti-doped HAp shows good affinity to organic compound which can be used for antibacterial applications (Tsuruoka et al. 2015). Samudrala et al. (2017) prepared borosilicate glassed doped with titanium dioxide (TiO2 ) to prepare compositions with precise degradation rate and superior biological response for bone-tissue engineering domain. The crystal structure of hydroxyapatite is hexagonal and when dopant ions replace the calcium ions, its crystal structure becomes distorted. The aim of this work was to prepare titania-substituted hydroxyapatite and to analyze their crystallographic behavior due to dopant and temperature.
2 Material and Method 2.1 Materials For the synthesis of titania doped and undoped hydroxyapatite, the chemicals which were used are: calcium hydroxide, (Ca(OH)2 , assay: 96%, Loba Chemie, India), orthophosphoric acid (H3 PO4 , assay: 85%, Loba Chemie, India), and titanium oxide (TiO2 , assay: 94%, Molychem, India). Synthesis of Pure and Titania Doped Hap. Hydroxyapatite was prepared using wet precipitation method as described elsewhere (Begam et al. 2017). Briefly, 0.6 M orthophosphoric acid was added dropwise for 2 h in 1.008 M calcium hydroxide solution. The chemical reaction was conducted in stirring condition at 80 °C and pH was 11. For doped hydroxyapatite, required amount of titanium oxide (5 wt%) was mixed into calcium hydroxide aqueous solution before addition of orthophosphoric acid solution. The prepared hydroxyapatite slurry was kept for aging for 24 h followed by filtration and washing with distilled water. After that, it was air dried at 80 °C for 24 h and dried cakes were sieved. The thus prepared powder is coded as HAp 80. The powder was calcined at 800 °C with holding time 2 h and heating rate 5 °C per min in a muffle furnace. Calcined powder was coded as HAp 800 °C. The calcined powder was then pressed at pressure of 150 MPa to make pellets using hydraulic
Effect of Temperature and Titania Doping …
279
press. Pressed pellets were sintered in a muffle furnace at 1250 °C with holding time 2 h and heating rate was 3 °C per min. The sintered samples were coded as HAp 1250 °C. Characterization. Both undoped and titania doped HAp were studied for phase analysis with X-ray diffractometry (XRD; Bruker diffractometer, D8 advance, Japan). The XRD scan was recorded in the 2θ range of 0–80° with scan speed of 1° per min. For calculation of lattice parameters (a axis and c axis), two peaks (3 0 0) and (0 0 2) was used. The formula for lattice parameter calculation is given below (Eq. 1) (Bhattacharjee et al. 2014). l2 4 h 2 + hk + k 2 1 + = d2 3 a2 c2
(1)
Here, d is the distance between adjacent planes in the set of Miller indices (h k l). The unit cell volume (V ) was calculated using the relation V = 2.589a2c (Bhattacharjee et al. 2014). The fraction of crystalline phase, i.e., the crystallinity (X c ) of the samples were also calculated by the following expression: V112/300 Xc = 1 − I300
(2)
Here, V112/300 is the intensity of hollow between planes (112) and (300). I300 is intensity of the (300) plane (Bhattacharjee et al. 2014). FTIR analysis was performed to study the bonds present in the material using KBr in UV-1800 Shimadzu instrument, Japan. The study was carried out in range of 400–4000 cm−1 .
3 Results and Discussion The XRD patterns of thus prepared powder(both pure and doped) are shown in Fig. 1a. In case of pure HAp 80, it was observed that the peaks were perfectly correlated with standard HAp with hexagonal crystal structure and P63/m space group, described in JCPDS data (card no. 09-0432). Major peaks were found at 2θ angle of 25.89° (0 0 2), 31.79° (2 1 1), 32.09° (3 0 0), 39.7° (3 1 0), 46.6° (2 2 2), and 49.36° (2 1 3). Whereas in doped HAp the peaks were little bit shifted. The characteristic peaks were observed at 2θ angle of 26.04° (0 0 2), 33.05° (3 0 0), 46.87° (2 2 2), and 49.69° (2 1 3). Both pure and doped calcined powders (Fig. 1b) were phase pure and wellmatched with XRD pattern of standard HAp with JCPDS file no 09-0432. The intensities of each peak were reduced in case of doped HAp compared to undoped. Peak shift was observed for doped powder at 300 (2θ angle of 33.03°). In case of sintered samples, both pure and doped HAp were phase pure and showed peak shift due to sintering at high temperature. Pure for HAp characterstics peaks were observed at 2θ angle of 26.30° (0 0 2), 32.21° (2 1 1), 33.37° (3 0 0), 32.61° (1
280
Y. Chopra et al.
Fig. 1 XRD spectra of pure and doped HAp at various temperature (a) at 80 °C, (b) at 800 °C, and (c) at 1250 °C
1 2), 47.13° (2 2 2), 49.89° (2 1 3). The peaks observed for sintered doped HAp were 2θ angle of 26.23° (0 0 2), 32.19° (2 1 1), 32.55° (1 1 2), 33.32° (3 0 0), 47.10° (2 2 2), and 49.86° (2 1 3). Table 1 shows the peak shifting due to effect of temperature and also due to dopant. Table 1 Characteristics peak shift due to temperature and doping Samples
Characteristic peaks 002
211
112
300
222
213
Pure HAp 80 °C
× √
× √
×
× √
×
×
×
×
Pure HAp 800 °C
×
×
×
×
×
Ti HAp 800 °C
× √
× √
× √
× √ √
× √
×
√
√
√
√
√
Ti HAp 80 °C
Pure HAp 1250 °C Ti HAp 1250 °C
×
× ×
Effect of Temperature and Titania Doping … Table 2 Lattice parameter of hydroxyapatite
281
Samples
a axis, Å
c axis, Å
Pure HAp 80 °C
9.650
6.8745
Ti HAp 80 °C
9.4193
6.835
Pure HAp 800 °C
9.4054
6.8563
Ti HAp 800 °C
9.3833
6.8485
Pure HAp 1250 °C
9.3039
6.7869
Ti HAp 1250 °C
9.2903
6.7691
The lattice parameter of hexagonal structure of hydroxyapatite crystal was calculated from XRD spectra as shown in Table 2. The a axis and c axis of doped samples were reduced compared to undoped samples. The size of titanium ion (ion radius 0.605 Å) is small compared to calcium ion (ion radius 0.99 Å). Due to doping, titanium replaces the calcium ions and it caused shrinkage in the crystal structure and therefore the lattice parameters of the crystal were decreased. Similar result was observed in literature in case of zinc doping (Begam et al. 2017). The lattice parameter also went on decrease due to an increase in temperature. As temperature increase, crystal size also decreases and hence also lattice parameters. Both doped and undoped HAp exhibited similar behavior on temperature. Similar observation was reported in literature (Gomes et al. 2012). Pure and zinc doped HAp showed reduced crystal size with temperature. The unit cell volume is presented in Fig. 2a. As the lattice parameter changed with temperature and doping, the unit cell volume was also changed. With increase in temperature there was loss of lattice H2 O from apatite lattice (Miyaji et al. 2005) and hence shrinkage of a and c axis was detected. The percentage of crystallinity was calculated from the XRD data by taking valley between peak (1 1 2) and peak (300) and the intensity of peak (3 0 0). The crystallinity
Fig. 2 a Unit cell volume and b fraction of crystallinity of pure and doped Hap
282
Y. Chopra et al.
(X c ) is the amount of crystalline phase within a material. The increase in intensity of the peaks with temperature also attributed the increase in percentage of crystallinity. It was observed that with temperature the percentage of crystallinity was increased as evidenced by the sharp diffraction peaks. There is vital role of dopant on crystallinity of HAp. As shown in Fig. 2b, the doped samples exhibited in decreased crystallinity compared to undoped samples at all temperature. As we seen, dopant causes shortening of lattice parameters and it causes disturbance in HAp crystallization (Begam et al. 2017). Stani´c et al. (2010) showed that copper and zinc dopant cause lessening of lattice parameters and also crystallinity. Figure 3 shows the FTIR spectra of all samples. The graph shows peaks for different functional groups present in the material. There was no peak shifting in doped HAp as depicted in FTIR spectra. The wide peak at 3571 cm−1 corresponds to the OH– group. Peaks for phosphate groups were found at 576, 567, 1008, and
Fig. 3 FTIR spectra of pure and doped HAp at different temperatures (a) at 80 °C, (b) at 800 °C, and (c) at 1250 °C
Effect of Temperature and Titania Doping …
283
1055 cm−1 . The peaks 1646 and 1420 cm−1 correspond to the carbonate groups. It was observed that with temperature the peak for hydroxyl group for calcined powder was sharp and very small. When the samples were sintered at 1250 °C that peaks were absent due to the evaporation of adsorbed OH bonds. This result was agreed with the XRD result as we seen reduction of lattice parameter with temperature.
4 Conclusion Bioactive calcium phosphate-based bioceramics are extensively used in bone tissue engineering due to their resemblance with the mineral phase of teeth and bones. They have outstanding biocompatibility, osteoconductivity, etc. In this work, a new type of bioceramics was prepared by doping titania in calcium phosphate. The prepared powder was heat treated at 800 °C and 1250 °C and the effect of temperature on crystal structure was studied. The crystallographic behavior of doped HAp was studied using XRD and FTIR. It was observed that the crystallinity was increased with temperature and decreased due to doping. Also, the lattice parameters were reduced due to doping as the titanium replaces by the calcium ions in the structure. Acknowledgments The authors are sincerely acknowledge the help and facilities provided by the Department of Biomedical Engineering, Department of Biotechnology, Department of Physics, and Department of Metallurgy, NIT Raipur, India for conducting this research work.
References Begam H, Kundu B, Chanda A, Nandi SK (2017) MG63 osteoblast cell response on Zn doped hydroxyapatite (HAp) with various surface features. Ceram Int 43(4):3752–3760 Bhattacharjee P, Begam H, Chanda A, Nandi SK (2014) Animal trial on zinc doped hydroxyapatite: a case study. J of Asian Ceram Soc 2(1):44–51 Bodhak S, Bose S, Bandyopadhyay A (2011) Bone cell–material interactions on metal-ion doped polarized hydroxyapatite. Mater Sci Eng, C 31(4):755–761 Geng Z, Wang R, Li Z, Cui Z, Zhu S, Liang Y, Liu Z (2016) Synthesis, characterization and biological evaluation of strontium/magnesium-co-substituted hydroxyapatite. J Biomater Appl 31(1):140–151 Gomes S, Nedelec JM, Renaudin G (2012) On the effect of temperature on the insertion of zinc into hydroxyapatite. Actabiomaterialia 8(3):1180–1189 Kalita SJ, Bhatt HA (2007) Nanocrystalline hydroxyapatite doped with magnesium and zinc: synthesis and characterization. Mater Sci Eng, C 27(4):837–848 Kalita SJ, Bhardwaj A, Bhatt HA (2007) Nanocrystalline calcium phosphate ceramics in biomedical engineering. Mater Sci Eng, C 27(3):441–449 Miyaji F, Kono Y, Suyama Y (2005) Formation and structure of zinc-substituted calcium hydroxyapatite. Mater Res Bull 40(2):209–220 Samudrala R, Penugurti V, Manavathi B (2017) Cytocompatibility studies of titania-doped calcium borosilicate bioactive glasses in-vitro. Mater Sci Eng, C 77:772–779
284
Y. Chopra et al.
Stani´c V, Dimitrijevi´c S, Anti´c-Stankovi´c J, Mitri´c M, Joki´c B, Ple´caš IB, Raiˇcevi´c S (2010) Synthesis, characterization and antimicrobial activity of copper and zinc-doped hydroxyapatite nanopowders. Appl Surf Sci 256(20):6083–6089 Tsuruoka A, Isobe T, Matsushita S, Wakamura M, Nakajima A (2015) Comparison of photocatalytic activity and surface friction force variation on Ti-doped hydroxyapatite and anatase under UV illumination. J Photochem Photobiol, A 311:160–165
Preparation and Characterization of Cellulose Nano Crystal/PVA/Chitosan Composite Film for Wound Healing Application Shubham Sen, Rashmi Agrawal, and Howa Begam
Abstract The aim of this study was to develop Cellulose Nano Crystal (CNC)-based sustainable biocomposite film. CNC was extracted from medical absorbent cotton using alkali and acid hydrolysis. The composite films were produced by reinforcing CNC in PVA/Chitosan polymer using solvent casting method. Films were characterized by XRD, contact angle, hemocompatibility, protein adsorption. CNC dispersed in PVA/Chitosan affects the surface properties and it enhanced the hydrophilicity of the composite film. The percentage of hemolysis of both composites was less than 5% which confirmed it as hemocompatible material. The percentage of protein adsorption of CNC-based composite was 39% higher compared to control. This result suggests that CNC is a good reinforcing material for biopolymer composite preparation. Keywords Cellulose nanocrystal · Biocomposite · Protein adsorption · Hemocompatibility
1 Introduction Transdermal drug delivery is a potential means to deliver active ingredients into the circulatory system through skin into blood because it is safe, effective, and comfort to patients (Shankar et al. 2018). Transdermal patches deliver drugs at a preset controlled rate to systemic circulation when applied to intact skin. It provides sustained and controlled delivery of molecules to body. Generally high dosage drug is administered into the patch and as the concentration of drug is higher in patch compared to blood, the drug will come into blood through diffusion process through skin for a long period that maintains constant drug concentration into blood (Ravichandiran S. Sen · R. Agrawal (B) · H. Begam Department of Biomedical Engineering, National Institute of Technology Raipur, Raipur, India e-mail: [email protected] H. Begam e-mail: [email protected] © Springer Nature Singapore Pte Ltd. 2021 A. A. Rizvanov et al. (eds.), Advances in Biomedical Engineering and Technology, Lecture Notes in Bioengineering, https://doi.org/10.1007/978-981-15-6329-4_25
285
286
S. Sen et al.
and Manivannan 2015). There are several biopolymers like gelatin, silk, starch, polyvinyl, alcohol, chitosan which are used to prepare transdermal wound healing patches (Khamrai et al. 2017). Cellulose is one of the most profuse biopolymer, which is commonly used as reinforcing elements for fabrication of composite biopolymer (Noshirvani et al. 2018). Cellulose is a polymer having glucose monomer, linked by β (1, 5) glycosidic bonds. Cellulose nanocrystal (CNC) is a promising biomaterials exhibiting many advantages like biocompatibility, mechanical properties, high water absorption capacity, low cost, renewability, etc., (Bajpai et al. 2017) and therefore it is attractive to use in bionanocomposites. CNCs has been used to prepare different types of biocomposites using polypropylene, polyvinyl alcohol, polyurethane, etc. Noshirvani et al. (2018) prepared CNC loaded starch/PVA nanocomposite film using solvent casting method for biodegradable packaging application. Bajpai et al. (2017) developed CNC-based chitosan films along with curcumin and silver nano particles for wound dressing application. They observed good results in terms of wound reduction in rat model. Wang et al. (2018) prepared antioxidant films using chitosan and epigallocatechin3-gallate with bacterial cellulose as reinforcement element. They observed bacterial cellulose improved tensile strength and reduced water solubility and it offered sustained release of drugs for long time. CNC and metallic nano particle-based composite nanomaterials showed good antibacterial properties against pathogenic bacteria (Perumal et al. 2018). Generally acid hydrolysis method is used for preparation of CNC (Singh et al. 2017). CNCs are available in different morphology depending on the source and method of extraction (George and Sabapathi 2015). Thambiraj and Shankaran (2017) extracted CNC from cotton with needle-shaped morphological structure. They did a preliminary experiments on fabrication of CNCbased film using cotton-based CNC. It is evident that there is no dearth of literature on extraction of CNC from cotton using hydrochloric acid, composite preparation, and its detailed characterization. The aim of this study was to develop CNC-reinforced wound dressing film using chitosan and polyvinyl alcohol. For this, CNC was isolated from medical absorbent cotton using hydrochloric acid hydrolysis. The efficiency of CNC-loaded film was compared with control (film without CNC). The water absorption, swelling properties, contact angles and protein adsorption studies were performed for both film with and without CNC.
2 Materials and Method 2.1 Materials The chemicals used for preparation of CNC and CNC composite films and for their characterization are given in Table 1
Preparation and Characterization of Cellulose …
287
Table 1 List of chemicals used for CNC extraction and film preparation Chemicals
Chemical formula
Purity
Company name
Hydrochloric acid
HCl
35–38%
Thomas Baker (Chemicals) Pvt. Ltd., India
Sodium hydroxide
NaOH
97%
SRL, India
Chitosan
C6 H11 NO4
75%
Himedia Laboratories, India
Polyvinyl alcohol
[CH2 CH(OH)]n
–
Loba Chemie, India
Bovine serum albumin (BSA)
–
98%
SRL, India
Ethanol
C2 H5 OH
99.9%
ChangshuHongsheng Fine Chemical Co.Ltd, China
Toluene
C7 H8
99.5%
SRL, India
Glacial acetic acid
CH3 COOH
99.5%
Loba Chemie, India
2.2 Preparation of Film Cellulose nanocrystals were extracted from medical absorbent cotton using alkali and acid hydrolysis as described in literature (Thambiraj and Shankaran 2017; AbuDanso et al. 2017). Briefly, cotton was carefully cleaned using distilled water and ethanol to eliminate any type of unwanted residues. After that, it was treated with 2% NaOH and acid hydrolysis was performed using hydrochloric acid. After that, CNC was centrifuged and washed with distilled water. Finally, it was dried at 60 °C for 12 h and the dried powder was used for film preparation. For the preparation of film solvent casting method was used. For this, 10 w/v% PVA solution was prepared using distilled water at 70 °C. 1 w/v% chitosan solution was prepared in distilled water containing acetic acid (1 v/v%) at 40 °C. After cooling, both solutions were mixed at the ratio of 70:30 in constant stirring followed by ultrasonication for one hour. For CNC loaded film, 3 wt% CNC powder according to dry weight of PVA and chitosan was added before ultrasonication. Both solutions were poured in glass Petri dish followed by drying at 40 °C for 12 h. PVA/Chitosan film was coded as film 1 and CNC/PVA/Chitosan composite was coded as film 2. The optical micrograph of as-prepared film is shown in Fig. 1. The color of both films was white and transparent.
2.3 Characterization Thickness. Thickness of composite films were measured using a micrometer screw gauge to nearest 0.01 mm. Thickness was measured at three random places of composite films and it was observed that thickness of films were same in each point of measurement. There was no difference between film 1 and film 2 and the thickness of each film was 1 mm.
288
S. Sen et al.
Fig. 1 Optical image of composite a film 1 and b film 2
X-Ray Diffraction (XRD). XRD of dried specimens was performed by X-ray diffractometry (XRD; Brukerdiffractometer, D8 advance, Japan) using Cu radiation with wavelength λ = 0.154 nm. The XRD scan was performed in range of 2θ angle of 0–80° and scan speed of 1° per min. Porosity. Porosity of the CNC films was determined by means of the liquid displacement method. The films were dipped in ethanol (99.9%) for 5–10mins and the wet weight was taken. The percentage of porosity was determined as, porosity = (WW − WD )/ρ × V where, WD and Ww are dry weight and wet weight of films, ρ = 0.789 g/cm3 (density of ethanol) and V is volume of films. Contact Angle. Contact angle was determined by conventional sessile drop technique (phoenix 300 instrument). Distilled water was dropped on the surface of the film and photo of drop and the film was captured by CCD camera. Protein Adsorption Studies. For protein adsorption studies, specimens of equal weight were immersed in ethanol (99.9%) for one hour and then washed thrice with PBS (pH 7.4) for 30 min. After that, the films were placed in glass test tube containing 2 ml of bovine serum albumin (BSA) (BSA conc. 1 mg/ml) for one hour. After one hour, the films were taken out and the solution was tested for unadsorbed proteins left in it. Proteins present in the solutions were estimated by folin-lowry method using BSA as standard. Absorbance (660 nm) was measured in UV-V is spectrophotometer (Shimazdu, Japan). Hemolysis Study. In vitro hemocompatibility test was conducted by hemolysis assays described in literature (Samanta and Chanda (2018)). Briefly, human blood
Preparation and Characterization of Cellulose …
289
was diluted in normal saline in 4:5 ratio. 2–3 g of films were taken in glass test tubes containing 10 ml saline and incubated at 37 °C for 30 min. After that, 0.2 ml diluted blood was mixed in both test tubes and again it was incubated for additional 60 min. Positive control and negative control were prepared by adding 0.2 ml diluted blood in 0.1% sodium bicarbonate (10 ml) and normal saline (10 ml), respectively. Both positive and negative controls were incubated at 37 °C for 60 min. After that, centrifugation (500 g) was carried for all test tubes for 5minutes, the supernatant was cautiously collected, and optical density was determined using spectrophotometer (Sl 177, ELICO, India) at 545 nm. The hemolysis (%) was determined using the following formula: %hemolysis =
O.D. of Test − O.D. of negative × 100 O.D. of positive − O.D. of negative
(1)
2.4 Statistical Analysis All data were presented as mean ± standard deviation (SD). All specimens were in triplicate (n = 3), unless stated otherwise. One way analysis of variance (ANOVA) with a Tukey’s post hoc test was carried out by using ORIGIN software. Significant differences were considered as: (p < 0.001; **p < 0.01; *p < 0.05).
3 Results and Discussion Thickness. The biocomposite films were prepared by using PVA/chitosan and PVA/Chitosan/CNC were uniform without any bubbles. The thickness of both samples measured as 1 mm. Thickness measurement is important because the physical properties of films depend on thickness of the film. There are no significant differences between two films. Porosity. The porosity of film 1 was calculated as 25.34 ± 17% and it was observed that after addition of CNC the porosity was reduced (17 ± 7.3%). The reduction of porosity of film 2 results in increase in apparent density. The density is beneficial for the mechanical property of scaffold. An optimum porosity is necessary for cell attachments and the movement of nutrients and it ultimately increases the implant’s bioactivity. For controlled release of drugs from scaffold, the porosity should be optimum without adversely affecting the structural or mechanical behavior of scaffold (Rambhia and Ma 2015). The release of drugs is generally occurs through diffusion process and after implantation scaffold exhibits strong burst release. Therefore, it is important to reduce initial burst release of drugs and maintain sustained release for longer period.
290
S. Sen et al.
Contact angle. Contact angle test is generally performed to determine the surface hydrophilicity of any biomaterial. Contact angle depends on the hydrophilic group, which is exposed on the surface of the material. The data of water contact angle of film 1 was 65.23° ± 0.54°. When CNC was added in PVA-chitosan-based films, it showed reduction in contact angle (54.53° ± 1.41°). The increase in hydrophilicity of CNC-based films may affect the cell adhesion, cell proliferation (Fig. 2). XRD study. The XRD spectra of both films are presented in Fig. 3. PVA displayed a semi-crystalline structure for hydrogen bond between the OH− groups of the PVA chains. Both films exhibited peak at 2θ angle of 20° for superimposed peak of chitosan and PVA (Amaral et al. 2013). Film 2 exhibited peak at 2θ angle of 22.74° for cellulose nanocrystal. Hemocompatibility study. Toxic materials can destroy erythrocytes and release hemoglobin. Hemocompatibility is considered one of key concerns in tissue engineering, particularly for scaffolds that will come into direct contact with blood. The percentage of hemolysis is important feature in assessment of biocompatibility of materials (Liu et al. 2014). The percentage of hemolysis of film 1 was 2.8 ± 0.49% and for film 2 it was 3.66 ± 0.19%. As both samples showed hemolysis less than 5%, therefore both films were highly hemocompatible according to the ASTM F756 standard (Samanta and Chanda 2018). The data suggested that CNC-based composite meets the requirement for medical biomaterials. Protein adsorption study. Protein adsorption is significant in controlling cell reaction to biomaterials (Saravanan et al. 2017). Biomaterials adsorb proteins on their surface to anchor osteoblast cells via integrins (Shankar et al. 2018). Initial protein adsorption is important because it determines the success of any biomaterials (Shankar et al. 2018). When a material is exposed to cell, proteins spontaneously adsorb on the surface of the materials. The adsorption of protein is mainly dependent on the surface chemistry, surface wettability, surface morphology, etc. (Le et al. 2013). In our study, we studied protein adsorptionability for one hour. It was observed that protein adsorption was higher in film 2 compared to the control (Figure 4). Inclusion of CNC in PVA/Chitosan film enhanced protein adsorption by 38%. As we found in contact angle analysis, film 2 showed more hydrophilicity and it can be related to enhanced protein adsorption rate.
Fig. 2 Water contact angle of a film 1 and b film 2 c Bar diagram showing contact angle difference between two films
Preparation and Characterization of Cellulose …
Fig. 3 XRD spectra of composite films Fig. 4 Bar diagram showing BSA protein adsorption of composite film after 1 h.
291
292
S. Sen et al.
4 Conclusion CNC was prepared from medical absorbent cotton in order to reinforce the PVA/Chitosan matrix with content 3 wt% by casting method. Films were uniform with no bubbles and the porosity of CNC composite film was reduced compared to PVA/chitosan matrix. Also, the hydrophilicity, hemocompatibility, and protein adsorptionability were enhanced due to CNC reinforcement. The improved proterties of CNC composite was mainly due to the inter-molecular hydrogen bonding interaction between CNC and polymer matrix. Acknowledgments The authors sincerely acknowledge the help and facilities provided by the Department of Biomedical Engineering, Department of Biotechnology, Department of chemical Engineering, Department of Physics, and Department of Metallurgy, NIT Raipur, India for conducting this research work.
References Abu-Danso E, Srivastava V, Sillanpää M, Bhatnagar A (2017) Pretreatment assisted synthesis and characterization of cellulose nanocrystals and cellulose nanofibers from absorbent cotton. Int J Biol Macromol 102:248–257 Amaral IF, Sousa SR, Neiva I, Marcos-Silva L, Kirkpatrick CJ, Barbosa MA, Pêgo AP (2013) Kinetics and isotherm of fibronectin adsorption to three-dimensional porous chitosan scaffolds explored by 125I-radiolabelling. Biomatter 3(2):e24791 Bajpai SK, Ahuja S, Chand N, Bajpai M (2017) Nano cellulose dispersed chitosan film with Ag NPs/Curcumin: An in vivo study on Albino Rats for wound dressing. Int J Biol Macromol 104:1012–1019 George J, Sabapathi SN (2015) Cellulose nanocrystals: synthesis, functional properties, and applications. Nanotechnol Sci Appl 8:45 Khamrai M, Banerjee SL, Kundu PP (2017) Modified bacterial cellulose based self-healable polyeloctrolyte film for wound dressing application. Carbohyd Polym 174:580–590 Le X, Poinern GEJ, Ali N, Berry CM, Fawcett D (2013) Engineering a biocompatible scaffold with either micrometre or nanometre scale surface topography for promoting protein adsorption and cellular response. Int J Biomater Liu Y, Cai D, Yang J, Wang Y, Zhang X, Yin S (2014) In vitro hemocompatibility evaluation of poly (4-hydroxybutyrate) scaffold. Int J Clin Exp Med 7(5):1233 Noshirvani N, Hong W, Ghanbarzadeh B, Fasihi H, Montazami R (2018) Study of cellulose nanocrystal doped starch-polyvinyl alcohol bionanocomposite films. Int J Biol Macromol 107:205–2074 Perumal AB, Sellamuthu PS, Nambiar RB, Sadiku ER (2018) Development of polyvinyl alcohol/chitosan bio-nanocomposite films reinforced with cellulose nanocrystals isolated from rice straw. Appl Surf Sci 449:591–602 Rambhia KJ, Ma PX (2015) Controlled drug release for tissue engineering. J Control Release 219:119–128 Ravichandiran V, Manivannan S (2015) Wound healing potential of transdermal patches containing bioactive fraction from the bark of Ficusracemose. Int J Pharm PharmSci 7(6):326–332 Samanta SK, Chanda A (2018) Study on the structure and properties of crystalline pure and Doped β-tri calcium phosphate ceramics. Mater Today: Proc 5(1):2330–2338
Preparation and Characterization of Cellulose …
293
Saravanan S, Chawla A, Vairamani M, Sastry TP, Subramanian KS, Selvamurugan N (2017) Scaffolds containing chitosan, gelatin and graphene oxide for bone tissue regeneration in vitro and in vivo. Int J Biol Macromol 104:1975–1985 Shankar S, Oun AA, Rhim JW (2018) Preparation of antimicrobial hybrid nano-materials using regenerated cellulose and metallic nanoparticles. Int J Biol Macromol 107:17–27 Singh S, Gaikwad KK, Park SI, Lee YS (2017) Microwave-assisted step reduced extraction of seaweed (Gelidiellaaceroso) cellulose nanocrystals. Int J Biol Macromol 99:506–510 Thambiraj S, Shankaran DR (2017) Preparation and physicochemical characterization of cellulose nanocrystals from industrial waste cotton. Appl Surf Sci 412:405–416 Wang X, Xie Y, Ge H, Chen L, Wang J, Zhang S, Feng X (2018) Physical properties and antioxidant capacity of chitosan/epigallocatechin-3-gallate films reinforced with nano-bacterial cellulose. Carbohyd Polym 179:207–220
Automated CAD System for Skin Lesion Diagnosis: A Review Lokesh Singh, Rekh Ram Janghel, and Satya Prakash Sahu
Abstract Skin cancer is deemed as the lethal type of cancer threatening worldwide with an increase in mortality rate per year. The growing incidences of melanoma skin cancer have introduced numerous treatment options. However, surgical treatment remains the basis for treating skin cancers. Automated skin cancer detection still remains a challenging task in the current scenario. Prior diagnosis of skin cancer requires computer-aided diagnosis. Computerized methods for analyzing images in dermoscopy are a subject of interest as dominant information regarding skin lesions can be retrieved. This paper has discussed the state of art techniques employed in CAD systems by providing domain facets of melanoma accompanied by efficient methods utilized in every phase. The phases comprise image preprocessing techniques, extraction, and selection of significant features, segmentation methods, and classification approaches for the identification of skin lesions. Inapplicability and future trends are discussed in the domain of research. Keywords Melanoma · CAD system · Segmentation
1 Introduction Cancer is a severe health issue and deemed as a major cause of an increase in mortality rate worldwide. American Cancer Society conducts a survey every year to evaluate the cancer statistics of the current year and archive the novel cancer instances, death rate, and survival of the United States. Society has estimated the occurrence of 600,920 deaths among 1,688,780 cases due to cancer incidences in 2017. According L. Singh (B) · R. R. Janghel · S. P. Sahu National Institute of Technology, G.E Road, Raipur 492001, India e-mail: [email protected] R. R. Janghel e-mail: [email protected] S. P. Sahu e-mail: [email protected] © Springer Nature Singapore Pte Ltd. 2021 A. A. Rizvanov et al. (eds.), Advances in Biomedical Engineering and Technology, Lecture Notes in Bioengineering, https://doi.org/10.1007/978-981-15-6329-4_26
295
296
L. Singh et al.
Table 1 Estimated new cancer cases and deaths by sex, the United States, 2017 (Siegel et al. 2017) Cancer cases
Estimated new cases
Estimated deaths
Both sexes
Male
Female
Both sexes
Male
Female
95,360
57,140
38,220
13,590
9250
4340
Melanoma of the 87,110 skin
52,170
34,940
9730
6380
3350
Other no epithelial skin
4970
3280
3860
2870
990
Skin (excluding basal and squamous)
8250
to the survey, cancer incidence rate and mortality rate are found around 20 and 40%, respectively, higher in men comparatively in women (Siegel et al. 2017). Based on the gender, evaluated cancer statistics of the United States for 2017 is presented in Table 1. Figure 1 graphically illustrates the estimated new cancer cases and death rate accordingly of three different categories of cancer of both sexes. It is clearly observed from Fig. 1 that the statistics of cancer cases and the death rate found to be higher in men than in women who are subjected to matter, as the statistics is fearsome and horrific. Skin cancer is one of the lethal forms of cancer considered today. As per the cancer statistics of Europe in 2012, 22,199 cases of deaths are reported among 100,339 cancer cases (Okur and Turkan 2018). Due to the higher death rate and excessive level of metastasis, melanoma is turning as one of the fatal types of cancer in the skin. According to one of the surveys conducted in 2008, melanoma found to be the 19th most mortal cancer worldwide due to its higher incidence rate with a projection of 200,000 new cancer cases in the northern countries (Oliveira et al. 2016a). Melanoma is a fatal disease that arises in pigment cells referred to as melanocytes (melanin-producing cells). Unlike other skin cancers, it simply dispersed above other Skin (excluding basal & squamous) Melanoma of the skin
ESTIMATED NEW CASES
Fig. 1 Evaluation of cancer cases and deaths
BOTH SEXES
4340 3350 990
FEMALE
9250 6380 2870
MALE
13590 9730 3860
38220 34940 3280
BOTH SEXES
57140 52170 4970
8250
95360 87110
CANCER CASES
Other nonepithelial skin
MALE
FEMALE
ESTIMATED DEATHS
Automated CAD System for Skin Lesion Diagnosis: A Review
297
Fig. 2 Melanoma stages (Okur and Turkan 2018)
tissues expeditiously. Cancer might be metastasized by the tissues, lymph system, or by the circulation of the blood. It gradually expands to the closer areas when expanded by tissues, but when it pierces through the lymph system or blood vessels, it expanded over the rest of the body tissues. This usually becomes an ill-posed problem when tissue turns cancerous as it expands through melanoma. Malignant melanoma can be successfully cured if detected at an early stage by a simple visual analysis, as its occurrence is clearly visible on the surface of skin. But the main practical problem that confronts us is that one can define the particular stage of melanoma only when a lesion has undergone surgery. The stage of melanoma can be identified using four primary features—(a) thickness of a tumor, (b) ulceration, (c) dispersion to the lymph nodes, and (d) dispersion to the rest of the body parts. The stages of melanoma are discussed in Fig. 2. Being aware of early detecting mechanisms of melanoma is the key to prevention from melanoma. Numerous techniques are available that have been developed for better estimation of skin lesions. A country like Turkey has found out the way to cope with this problem by running 19 clinics for investigation of melanoma. Several facilities are now available online for taking appointments in numerous health centers worldwide. ‘ABCDE of melanoma’ serves as a guideline for increasing the awareness regarding skin lesions, where ‘A’ stands for Asymmetry, ‘B’ for Border irregularity, ‘C’ for distinct Colors, ‘D’ for Diameter features of lesions, and ‘E’ represents Evolution which acts as an indicator of the rapid growth of lesions. An individual investigated suspected as per the above-mentioned guidelines accordingly must reach to the experts for clinical examination. This problem can be investigated visually by experts. This phenomenon has been widely observed by investigators with an average inspection accuracy of 65%, which is relatively a low accuracy that has emerged from
298
L. Singh et al.
the imperfection of the examiners. Automated detection systems are thus pivotal for better investigation of skin cancer. Digitized image processing approaches are therefore required which results in significant facts about skin cancer which would be of great importance for the experts (Okur and Turkan 2018).
2 Skin Cancer It is of great interest to know about skin as the biggest organ of the body structure, which covers approx. 16% of the body mass. Skin structure is composed of three layers, depicted in Fig. 3 are, namely epidermis layer, dermis layer, and the subcutaneous fat layer which further composed of numerous components like epithelial, mesenchymal, glandular, and neurovascular. Epidermis, the first layer is the thinnest and outermost layer of the skin surface which can clearly be observed by the naked eye. It highly protects the body from being damaged from chemicals, ultraviolet rays, etc. This layer serves as a protective shield for the skin surface. Dermis, the second layer is responsible for sweat generation, hair growth, and manages the flow of blood to the skin. Subcutaneous fat layer, the third and bottom layer of the skin, is responsible for connecting the dermis to the body’s muscles and bones and manages the body temperature (D’Orazio et al. 2013; Mark Elwood and Jopson 1997; Lowe 2006; Slominski and Wortsman 2013; Borja-Cacho and Matthews 2008; Slominski et al. 2012). Cancer begins when the growth of cells becomes out of control in the body. Cells can turn cancerous in any part of the body and can spread to the rest of the body areas. Melanoma skin cancer begins in specific types of cell. Figure 3 depicts the three main types of cells in the top layer of the skin (called the epidermis) (Fig. 4). Squamous cells: These cells are flat and thin by structure and observed in the tissue which constitutes the outer skin layer. Basal cells: These cells are small in size and round in shape usually found in the lower section of the epidermis. These cells continuously separate themselves for the formation of new cells, so squamous cells can be replaced that constitute the skin’s surface. Melanocytes: These cells can turn into melanoma. They constitute a browncolored pigment referred to as melanin which turns the skin brownish. This pigment prevents the deep layers of the skin from being harmed by sun exposure. More amount of pigment is generated by the melanocytes when skin comes in contact with the sun turning the skin darker (American Cancer Society 2017).
Automated CAD System for Skin Lesion Diagnosis: A Review
299
Fig. 3 Types of skin layer and skin cells (Korotkov and Garcia 2012)
2.1 Melanoma Skin Cancer The unusual development of the melanocytes is responsible for the formation of malignant tumors usually referred to as melanoma (Pathan et al. 2018), which is a type of cancer that usually starts in melanocytes. Malignant melanoma is also called cutaneous melanoma (Melanoma et al. 2007). Melanoma tumors are brown or black in color when melanin produced by the melanoma cells, but tumors appear pink or white in color when melanoma cells stop producing melanin. Melanoma can occur in any part of the skin surface but usually begins on the chest and backside in males and on legs in females. Neck and face are other common parts. Skin with dark pigments are at low risk of having melanoma at these common parts but can be risky at other parts like palm of hands, soles of the feet, and under the nails. Melanoma occurs more frequently at skin than at the rest of the body parts like eyes, mouth, genitals, etc. Melanoma occurs less commonly compared to the BCC (Basal Cell Cancer) and SCC (Squamous Cell Cancer) but is highly hazardous as it spreads speedily to other body parts if left undetected early (American Cancer Society 2017). Melanoma can’t be cured when reached to an advanced stage and possible treatment involves surgery, immunotherapy, chemotherapy, and/or radiation therapy (Society 2011; Kaufman 2005; Jerant et al. 2000).
300
L. Singh et al.
2.2 Non-Melanoma Skin Cancer Skin cancer, which doesn’t fall under the category of melanoma is generally referred to as non-melanoma skin cancer as their growth depends on skin cells rather than melanocytes. Their nature is entirely different from those of the melanomas and essentially requires different treatment (Kaufman 2005). Basal cell carcinoma. As per the experts, around eight skin cancers out of ten are found to be BCC. This cancer usually begins to grow on the areas, which are being harmed due to sun exposure in particular to the head and the neck. The development of BCC is very slow and very rarely spread to the rest of the body parts. It can be hazardous when not treated properly, in such cases, it spreads and starts developing in the rest of the body parts and harms the bones and other tissues by attacking underneath the skin. If not cured properly, it might occur again after some time on the same place over the skin surface. Individuals suffering from the BCC are at high risk as it might spread to other places in the body (American Cancer Society 2017). Squamous cell carcinoma Heading. This cancer is less hazardous than the BCC as per the fact, where two individuals found to be suffering from the SCC out of ten. SCC generally occurs on such areas of the body which are open to sun like face, ears, neck, and lips (American Cancer Society 2017; Kaufman 2005; Jerant et al. 2000). Keratoacanthomas. Tumors of dome-like shape originated due to sun exposure are termed as keratoacanthomas. They evolve very quickly but later grows slowly. Some of them are not deemed hazardous as they get might be cured automatically with time without any treatment. But, some of them sustain to increase and few of them might spread to the rest of the body parts. The growth of Keratoacanthomas is difficult to examine by the experts which results in improper prediction which is mistakenly considered as SCC by the experts (American Cancer Society 2017). Merkel cell carcinoma. Merkel cells are neuroendocrine cells that are responsible for hormone making cells. These cells originate over the base of the top skin layer— epidermis. Markel cells are found to be closer to the nerve endings in the skin. They are responsible for sensations of being touched. This cancer begins with the uncontrollable growth of Markel cells. Markel cell carcinoma is also known as neuroendocrine carcinoma as Markel cells are a kind of neuroendocrine cells (American Cancer Society 2017; Jerant et al. 2000). Kaposi sarcoma. Kaposi sarcoma is a type of cancer in which abnormal cells generate purple, red, or brown color tumors over the skin surface, termed as lesions. Apart from the skin surface, they can also evolve in the rest of the body parts like lymph nodes, lungs, etc. These skin lesions due to Kaposi sarcoma often appear over the legs or the face and cause the affected area to swell painfully. This cancer becomes lethal when lesions grow in lungs, liver, or digestive tract (American Cancer Society 2017).
Automated CAD System for Skin Lesion Diagnosis: A Review
301
Lymphoma of the Skin. Lymphoma is a type of cancer that starts growing in white blood cells lies in the immune system of the body. When it starts growing in the skin, it is termed as skin lymphoma or cutaneous lymphoma. Basically, it is of two types: • Hodgkin lymphoma: it is also termed as Hodgkin’s lymphoma or Hodgkin’s disease • Non-Hodgkin lymphoma: it is also termed as non-Hodgkin’s lymphoma or NHL (Kaufman 2005; Jerant et al. 2000).
3 Image Acquisition Cameras and digital videos are the sources to acquire clinical images for the acquisition of the images. Images obtained from mobile distances or acquired at distinct lightning conditions make the imaging inconsistent. The problem arises when the lesion is small in size due to the poor resolution of the image. The presence of artifacts like hairs, reflections, shadows, and skin lines are other hitches in image acquisition which obstruct in analyzing the skin lesions. Indeed, a non-invasive method referred to as ELM (Epiluminescence Microscopy) is employed for acquiring the images. In this method, the lesion is engrossed in oil, and images are obtained with the help of dermatoscopic devices comprises of a camera (Pereira et al. 2016). A miscellaneous dataset is the key requirement in the development of a robust CAD system for the better detection of melanoma. Table 2 illustrates the dermatological datasets which are publically available. It is of great interest to know about skin as the biggest organ of the body structure which covers approx. 16% of the body mass. Skin structure is composed of three layers,
302
L. Singh et al.
Fig. 4 Examples of dermoscopy (a and c) and macroscopic (b and d) images: a and b are images of melanoma in situ, and c and d are of invasive melanoma (these images are publicly available in (Oliveira et al. 2016a)
4 CAD System 4.1 Preprocessing For the betterment of the quality of images, preprocessing is deemed as the initial stage of the diagnosing system where the unwanted noise such as bubbles, hair, etc., is removed as they might result in inaccurate classification. Various causes are responsible for the preprocessing of the source image like (a). low contrast amidst skin lesion and surrounding skin, (b). irrelevant borders, (c). Artifacts such as skin lines, hairs, black frames, etc., which in turn might affect the accuracy. A clear demonstration of the given input image before and after hair removal is illustrated in Fig. 5. Image enhancement, image restoration, and hair removal are the key steps of image pre-processing and are clearly illustrated in Fig. 6 (Mehta and Shah 2016).
Automated CAD System for Skin Lesion Diagnosis: A Review
303
Table 2 Dataset details Datasets
Number of images
Number of lesion images
Dermoscopic and lesion feature annotation
Reference
ISIC Archive (2016 ISBI challenge)
900
273—ME 627—Non-ME
Globules Streaks
Codella et al. (2018)
ISIC Archive (2016 ISBI challenge)
2000
374—ME 254—Seborrheic keratosis 1372—Benign Nevi
N/W Negative N/W Milla like cysts streaks
Berseth (2017)
PH2
200
80—Common Nevi 40— ME 80—Atypical nevi
Asymmetry Colors Pigment N/W Dots Globules Streaks Regression Areas Blue—White veil
Mendonca et al. (2013)
Dermatology Atlas
8084
80—ME
NO
Dermoscopy Atlas (https://www.dermos copyatlas.com)
Dermnet Skin Disease Atlas
23,000
190—ME
NO
Atlas of Dermatology (https://www.der mnet.com/)
EDRA Interactive Atlas of Dermoscopy
NO
Invasive Pigment N/W melanomas Irregular Dots in situ Globules melanoma, Spitz nevi, Clark nevi, Bowen disease (80— ME, 120—Benign)
Argenziano et al. (2000)
Dermofit 1300 Image Library
76—ME
NO
Dermofit Image Library—Edinburgh Innovations (https:// licensing.eri.ed.ac.uk/ i/software/dermofitimage-library.html)
Dermquest
NO
308—ME
Type of ME
Search results (https://www.dermqu est.com/results/?q= Malignantmelanoma)
Dermis
NO
300—ME
NO
DermIS.https://www. dermis.net/dermis root/en/home/index. htm)
* ME
= Melanomas, N/W = Network
304
L. Singh et al.
Fig. 5 a Original image b hair removal.
Fig. 6 Preprocessing techniques
4.2 Lesion Segmentation Segmentation is the primary method for the separation of data into a significant region of interest (ROI). The quality of the segmented image is of maximal importance as it directly affects the performance of the CAD system. The lesion is segmented when the boundary of a lesion is delineated from the given input image with the help of automated image segmentation mechanisms (Rodríguez and Sossa 2017). Figure 7 clearly illustrates the image before and after segmentation. IMAGEJ software can also be employed for the purpose of colored segmentation (Jaleel and Salim 2013) (Figs. 8, 9 and 10).
Fig. 7 a Original image b segmented image (Jaleel and Salim 2013)
Automated CAD System for Skin Lesion Diagnosis: A Review
305
Fig. 8 Image segmentation techniques
Based on several image attributes like histogram, color, gradient, wavelet coefficients, etc., block-based methods separate the image into rectangular blocks. Blockbased segmentation methods are further categorized into three types as per the fundamental traits of pixels (Jaglan et al. 2019). Region-Based Detection Method. In region-based detection methods, regions are generally analyzed by the center of gravity, which is inconvertible corresponding to the rotation, scaling, and skewing and remains constant under random noise and gray-level variation. Segmentation approaches play a key role in diagnosing the region-based features. The accuracy obtained after segmentation notably affects the resulting registration. The image segmentation is performed, respectively, in an iterative manner with the registration wherein at each step segmentation parameters are tuned (Zitová and Flusser 2003). Boundary-Based Detection Methods. The limitations left uncovered by the regionbased approaches are overcome using boundary-based approaches for segmentation and classification. The segmentation of image is performed into regions through unseen changes occurred in the intensity of an image. Boundary based methods are categorized into two, namely ridge detection and edge detection. The ridge detection fulfills the aim by capturing the major axis of symmetry of an enlarged object, while the edge detection fulfills the aim by capturing the borders of the object (Jaglan et al. 2019; Pathan et al. 2018).
306
L. Singh et al.
4.3 Feature Extraction In computerized PSL analysis, in order to classify a lesion most automated systems aim to extract features from the images and represent them in a way that can be understood by a computer. Features regarding skin lesion are divided into three types: shape-type features, color features, and features based on texture. Further, features are then extracted to diagnose the patterns and detect skin lesions using macroscopic and dermoscopic images. In CAD system, feature extraction is another stage next to the segmentation stage for performance improvement (Oliveira et al. 2016a; Korotkov and Garcia 2012). ABCD Rule. This method distinguishes benign and malignant melanoma (Ali et al. 2014). This approach relies on a semi-quantitative analysis of the criteria, namely Asymmetry—A, Border—B, Color—C, Different dermoscopic structures—D (Leo et al. 2009; Singh et al. 2014). Menzies Method. Menzies’s approach basically divides the feature into two categories. The first category is the positive feature group and the second category is the negative feature group. The positive and negative features are discussed in Table 3 (Okur and Turkan 2018). Seven Point Checklist. This checklist comprises 7 distinct features which when combined act as an indicator of melanoma. Likewise, the ABCD rule, each feature is labeled with a value that is calculated for the analyzed lesion based on existing specific features (Okur and Turkan 2018). Cash Algorithm. CASH algorithm stands for Color, Architecture, Symmetry, and Homogeneity/Heterogeneity. Feature architecture included by this algorithm is not employed by other scoring algorithms. The architecture indicates the structure of the lesion in terms of dermoscopic structure and colors (Pathan et al. 2018). Table 3 Menzies’s scoring
Negative features
Positivef
• Symmetry of lesion
• Blue-White veil
• Presence of a single color
• Multiple brown dots • Pseudopods • Radial Streaming • Scar-like depigmentation • Globules • Multiple 5–6 colors • Multiple blue-gray dots • Broadened network
Automated CAD System for Skin Lesion Diagnosis: A Review
307
Shape Features. The organization of an object defines the shape of a particular object which is represented using boundary, region, etc. The investigation of asymmetry features is performed by separating the lesion’s region into two sub-regions by an axis of symmetry for analyzing the similarity of the area (Chang et al. 2005). Color Features. The contrast of the skin lesions is represented by RGB color space. To acquire more information regarding color of the lesion some other color spaces have been utilized, namely RGB, HSV, HVC, CMY, YUV, I1/2/3, L*C*H, CIEXYZ, CIELAB, and CIELUV (Oliveira et al. 2016a). Texture Features. Basically, texture features can be explained by characteristics like coarseness, contrast, and directionality. For the human visual system, the texture is identical (Lew 2013). Texture feature extraction comprises three main classes, namely Statistical, structural, and transform-based methodologies (Haralick et al. 1973).
4.4 Feature Analysis and Selection Feature selection is the process of searching the most relevant and significant features among all features. An advantage of employing feature selection methods is the reduced computation time and improved prediction performance in the field of machine learning (Chandrashekar and Sahin 2014). A large number of features, when provided as an input to the classifier, might result in redundancy and increased computational complexity. Therefore, to improve the robustness of the classification model, feature selection is the next step after feature extraction which enhances the classification performance. A significant facet of feature selection is maximal relevancy with less redundancy (Singh et al. 2017). Filter Approach. No learning methodology is opted by filter-based methods for the selection of the best subset of features. Dominant features are evaluated based on ‘scoring’ criteria and highly scored features are then chosen for further classification process which makes filter-based methods faster than the wrapper based methods. Following are the filter-based methods described: (A) χ2 Test The chi-square test, also represented as χ2 test or known as Pearson’s chisquare test, demonstrates an apparent difference among expected frequencies and observed frequencies. The following equations represents statistical independence between two events x and y. P(x y) = P(x)P(y) or
(1)
P(x/y) = P(x) and P(y/x) = P(y)
(2)
308
L. Singh et al.
(B) Euclidian Distance Euclidean distance can be employed to measure the distance among pair of instances p and q in an N-dimensional feature space. Usually, it is evaluated as the square root of the addition of the squared difference among the related coordinates. d(a, b) = {i(ai − bi)2}1/2
(3)
(C) Information Gain Information gain evaluates the amount of information a feature provides regarding the class. It describes the relevancy of an attribute. It indicates the percentage of separation of data by the attribute according to the classification. The entropy can be calculated as follows: Entropy(s) =
n
−Pi log2Pi
(4)
i=1
where n = number of classes, Pi = probability of s belongs to class i, (Vanaja 2014). Wrapper Method. Wrapper-based approaches employ learning methods for the performance evaluation of classification algorithms. Wrapper approaches are better than filter methods as they provide higher classification accuracy but at a cost of higher computational complexity. • Forward selection: The process begins with a null set and at a time one feature is added, the rest of the features are then appended to the existing subset for the evaluation of the novel subset. • Backward elimination: The process begins with a featured set comprises of all features and features are removed one at a time. • Forward stepwise Selection: The process begins with a null set and features are appended and removed one at a time (Cadenas et al. 2013). Embedded Method. Embedded approaches incorporate variable elimination during the training process which is specific to the employed learning methodologies. This approach interacts directly with the classification methods. It is faster than the wrapper methods. Following are the few embedded feature selection methods: • Decision Trees. • Weighted Naïve Bayes. • Variable elimination using the weighted vector of SVM. Hybrid Approach. Hybrid method is an amalgamation of filter and wrapper approaches. As a filter-based method candidate, feature subset is chosen from the original feature subset which is then refined by the wrapper based methods. It utilizes the benefits of both of the two methods (Vanaja 2014).
Automated CAD System for Skin Lesion Diagnosis: A Review
309
4.5 Classification Classification comprises the last stage of CAD system to decide the Pigmented Skin Lesion class. A brief description of several classification approaches employed in diagnosing melanoma is explained in this section (Giotis et al. 2015). SVM. The support vector machine follows the principal of structural risk minimization with an aim to search the classification method which curtails the boundary of the expected error (Gehler and Schölkopf 2009). It searches for the optimal separating hyperplane which gives the maximum margin and seeks for the closest data point of the training dataset amidst 2 classes (Codella 2015). K-NN. K-NN referred to as K-nearest neighbor is one of the simplest machine learning methods. It works on the principle of feature similarity. A very minimum training phase is required which makes the training process faster. It stores the complete available data points and classifies novel data points using the distance function like Euclidean, Manhattan, and Minkowski functions. Data samples are classified using the majority vote of their neighbors (Rourrur and Rvise 1982; Erickson et al. 2017). Discriminant Analysis. Discriminant analysis is used for classification predictive problems and is defined by Fisher (1936), where two or more samples or populations are defined a priori, and one or more than one novel sample is then classified into one of the defined samples. It defines the relationship between dependent variables and independent variables (Analysis 2018). ANN. ANN stands for Artificial Neural Network, also referred to as neural network. It models the complex relationships between input and output. Pattern recognition is the basic task performed by ANN in the field of medical imaging. It is a computational model which relies on the framework of a biological neural network. Information passes from the network directly affects the organization of the ANN as neural network learns on the basis of input and output (Ercal et al. 1994). Decision Tree. A decision tree is a supervised machine learning algorithm employed in classification problems. A decision tree is a group of questions arranged in a hierarchically structured form and is depicted graphically in tree form. It follows a branching method to demonstrate each possible output of a decision. Decision trees are designed by evaluating a group of training data points for known class labels which are then used for classifying prior unseen samples (Yu et al. 2016). Naïve Bayes. Naïve Bayes is one of the most efficacious classification algorithms of machine learning. Naïve Bayes classifiers are a collection of probabilistic classifiers that rely on Bayes theorem with an assumption among features. It is not an individual algorithm but a collection of several methods works on the same principle i.e. each classified feature does not depend on each other. It is easy to construct with a simple evaluation and thus useful for big datasets (Lee et al. 2011).
310
L. Singh et al.
Fig. 9 Linear classifier in two-dimensional space
Fig. 10 Illustration of a 3-NN rule in a three-dimensional pattern space
Random Forest. Random forest is a supervised machine learning approach used for both classification and regression tasks. It constructs forests and makes them random. The forest it constructs is a decision trees ensemble usually trained with bagging approach. Rather than seeking the significant or dominant features during the splitting of a node, it seeks the most significant feature from a random subset of features (Chan and Paelinckx 2008).
5 Performance of Evaluation Measures Evaluation methods are the key elements in assessing the performance of classification models. The designed classification model might give satisfactory results when evaluated using any one of the metrics but provides unsatisfactory results when evaluated using other metrics. Thus, assessment of the classification model using a single evaluation measure is not sufficient. Table 4 describes several assessment methods for evaluating the effectiveness of the designed model (Pathan et al. 2018; Tharwat 2018).
Yr
2017
2017
2016
2016
2016
2015
2015
References
Yu et al. 2016)
Dalila et al. 2017)
Kasmi and Mokrani 2016)
Oliveira et al. 2016b)
Abbas et al. 2016)
Amelard et al. 2015)
Abuzaghleh et al. 2015)
Geodesic active contour
Ant Colony Based Segmentation
Fully Convolutional Residual Network
Segmentation
200 dermoscopic
206 dermoscopic
LIBSVM
Otsu thresholding + (NO) Active Contour using Sparse- Field level - set method
MV-SVM
Support Vector Machine
TDS
KNN, ANN
Deep Residual Network
Classification Models
LIBSVM
(NO)
(NO)
(NO)
Relief Algorithm
Low level feature algorithm
Feature Selection
(NO)
(NO)
350 dermoscopic Circular center of each PSL
408 dermoscopic Chan-Vese
200 dermoscopic
172 dermoscopic
1250 dermoscopic
Type of Image
Table 4 Summarized analysis of classification approaches employed in the computerized analysis
M/B/Atypical
M/B
M/B
M/B
M/B
M/B
M/B
Classification
(continued)
Benign = 96.4%, Atypical = 95.8%, Melanoma = 97.6%
Accuracy: 83.58%, 81.37%, 81.18%
Accuracy: 93.1%, Sensitivity: 94.1%, Specificity: 84.1%
Accuracy: 74.37%
Accuracy: 94.1, Sensitivity: 91.3%, Specificity: 95.9%
Accuracy: 85.23% for KNN, Accuracy: 93.63% for ANN
Accuracy: 85.6% Sensitivity: 54.8%, Specificity: 93.2%
Evaluated Results
Automated CAD System for Skin Lesion Diagnosis: A Review 311
Yr
2015
2015
2015
2015
References
Codella 2015)
Giotis et al. 2015)
Shimizu et al. 2015)
Rastgoo et al. 2015)
Table 4 (continued)
180 dermoscopic
968 dermoscopic
170 dermoscopic
200 dermoscopic
Type of Image
Threshold
Threshold
K-MEANS clustering
(NO)
Segmentation
PCA (NO/NO)
Incremental Stepwise (828/25)
(NO)
(NO)
Feature Selection
Random Forest
Linear Classifier
Ensemble method (CLAM, CIA-LVQ,Naïve Bayes
Support Vector Machine
Classification Models
Melanoma/ dysplastic nevus
M/N
M/N
M/B
Classification
(continued)
Sensitivity: 98.0%, Specificity: 70.1%
Melanoma: Detection Rate: 90.47%; Nevus: Detection Rate: 82.52%, bcc: 82.60%Detection Rate sk: 80.62%Detection Rate
Accuracy: 81.0%, Precision: 0.741%, Negative Predicted Value: 85.8%
Accuracy: 91.1%, Sensitivity: 97.1%, Specificity: 65.1%, Precision: 92.1%., F-measure: 94.2%, Auc: 95.2%
Evaluated Results
312 L. Singh et al.
Yr
2015
2015
2015
2014
2014
2013
References
Barata et al. 2015)
Møllersen et al. (2017)
Amelard et al. 2015)
Zortea et al. 2014)
Schaefer et al. 2014)
Abbas et al. 2013a)
Table 4 (continued)
120 dermoscopic
564 dermoscopic
206 dermoscopic
206 macroscopic
210 dermoscopic
Database1: 200 Database2: 482 dermoscopic
Type of Image
Sequential Forward Select (53/7.6) Fast Correlation Based Filter (437/74) Sequential Floating Forward Selection
Threshold region = growing & merging Threshold + Improved Dynamic Programming
– (62/–)
Wrapper & Filter (59/19)
Fusion Strategy (NO/NP)
Feature Selection
(NO)
(NO)
(NO)
(NO)
Segmentation
Support Vector Machine
Support Vector Machine
DA
Support Vector Machine
DA
Random Forest
Classification Models
M/N
M/B
M/B
M/B
Classification
M/N
Classification
(continued)
Sensitivity: 88.3%, Specificity: 91.4%
Accuracy: 93.84%, Sensitivity: 93.75%, Specificity: 93.83%
Sensitivity: 86.1%, Specificity: 52.3%, Correct Rate: 63.2%
Accuracy: 83.58%, Sensitivity: 91.02%, Specificity: 73.46%
Correct Rate: 81.2%, Sensitivity: 83.0%, Specificity: 80.4%
Sensitivity: DB1: 98.0%; Specificity: 90.1%; Sensitivity: DB2: 83.3%, Specificity: 76.2%
Evaluated Results
Automated CAD System for Skin Lesion Diagnosis: A Review 313
134 macroscopic
152 macroscopic
Ma and Staunton 2013 2013)
2013
2013
2013
2012
Cavalcanti et al. 2013)
Barata et al. 2014)
Abbas et al. 2013a)
Garnavi et al. 2012)
289 dermoscopic
120 dermoscopic
176 dermoscopic
350 dermoscopic
2013
Abbas et al. 2013b)
Type of Image
Yr
References
Table 4 (continued)
Gain Ratio method SVM + RF + LMT + HNB classifiers
Global threshold + adaptive histogram threshold
Support Vector Machine
AdaBoost
stage one: KNN; stage two: maximum likelihood
ANN
ML-SVM, ML-KNN, AdaBoost MC
Classification Models
SFFS (NM/NM)
Individual and combined feature analysis
(stage 1: 52; stage 2: 12)
Correlation Analysis (25/13)
PCA
Feature Selection
Dynamic programming
Threshold
Threshold
(NO)
450 * 450 ROI is selected
Segmentation
M/B
M/N
M/N
M/B
M/B
Reticular, Globular, Cobblestone, Homogeneous, Parallel ridge, Starburst, Multicomponent
Classification
(continued)
Accuracy: 91.27%, Auc: 93.8%
Melanoma: Sensitivity:88.2%, Specificity:91.20%, Auc: 88.1%; Nevus: Sensitivity:86.6%, Specificity:88.3%, Auc:82.5%
Sensitivity:96.2%, Specificity:80.1%
Accuracy: 99.35%, Sensitivity:100%, Specificity:97.79%
Sensitivity: 83.0%, Specificity: 90.1%, Auc: 89.2%
Sensitivity: 89.3%, Specificity: 93.76%
Evaluated Results
314 L. Singh et al.
2012
Garnavi et al. 2012)
289 dermoscopic
Type of Image Threshold
Segmentation GRFS (35,455/23
Feature Selection Random Forest
Classification Models M/B
Classification
Accuracy: 91.27%, Auc: 93.8%
Evaluated Results
where M/N = Melanoma/Nevus, DA = Discriminant analysis, acc = Accuracy, se = Sensitivity, sp = Specificity, ppv = Positive Predicted Value (Precision), npv = Negative Predicted Value, fm = F = Measure, cr = Correct Rate, dr = Detection Rate, Auc = Area Under the ROC Curve, bcc = Basal Cell Carcinoma, sk = Seborrheic Keratosis, M/B = Melanoma/Benign
Yr
References
Table 4 (continued)
Automated CAD System for Skin Lesion Diagnosis: A Review 315
316
L. Singh et al.
Table 5 Classifier evaluation parameters (Tharwat 2018) Measures
Formula
ACC
T N +T P F N +F P+T N +T P
SE (TPR)
T PR + T NR − 1
Ratio of correctly classified positive instances
SP (TNR)
TN F P+T N TP T P+F P
Ratio of correctly classified negative instances
NPV
TN T N +F N
Ratio of correctly classified negative instances out of total negative predictive instances
FPR
1 - TNR
Complement of specificity
FNR
1-TPR
Complement of sensitivity
YI
T PR + T NR − 1
It combines the TPR and TNR into a single measure
FM
2T P 2T P+F P+F N
Harmonic mean of PPV and TPR
PPV
GM
√
T PR ∗ T NR
Description Proportion of correctly classified instances out of total instances
Ratio of correctly classified positive samples out of total positive predictive instances
Evaluation of sensitivity and specificity
where ACC = Accuracy, SE = Sensitivity, SP = Specificity, PPV = Positive Predicted Value (Precision), NPV = Negative Predicted Value, FPR = False Positive Rate, FNR = False Negative Rate, YI = Youden’s Index, FM = F-Measure, and GM = Geometric-mean
6 Conclusion and Future Trends This study represents an overview of the research conducted in the computerized analysis of dermatological images. The aim of this paper is to describe all the stages of a CAD system for the detection of skin cancer using several classification approaches. This review gives key insights into the recent developments in computerized analysis of Pigmented Skin Lesions by employing dermoscopic images. A thorough evaluation of cancer statistics constitutes of the domain facets is discussed. Recent evolving computational approaches for segmentation and feature extraction are elaborated in brief. Additionally, classifiers for the identification of skin lesions are discussed along with performance evaluation measures. This review concludes various methods focusing on the classification of skin lesions employed in automated computer-Aided Diagnosis system. This analysis proves to be helpful for dermatologists in diagnosing skin cancers. Though numerous researches have been conducted addressing detection and prevention methods regarding skin cancer still, novel methods might need to be addressed and designed to fill the research gaps for performance improvement of the CAD systems (Table 5).
Automated CAD System for Skin Lesion Diagnosis: A Review
317
References Abbas Q, Celebi ME, Serrano C, et al (2013b) Pattern classification of dermoscopy images: a perceptually uniform model. Pattern Recognit 46:86–97. https://doi.org/10.1016/j.patcog.2012. 07.027 Abbas Q, Emre Celebi M, Garcia IF, Ahmad W (2013a) Melanoma recognition framework based on expert definition of ABCD for dermoscopic images. Ski Res Technol 19:1–10. https://doi.org/ 10.1111/j.1600-0846.2012.00614.x Abbas Q, Sadaf M, Akram A (2016) Prediction of dermoscopy patterns for recognition of both melanocytic and non-melanocytic skin lesions. https://doi.org/10.3390/computers5030013 Abuzaghleh O, Member S, Barkana BD (2015) Noninvasive real-time automated skin lesion analysis system for melanoma early detection and prevention. 3 Ali A, Couceiro MS, Hassenian AE (2014) Melanoma detection using fuzzy c-means clustering coupled with mathematical morphology. 73–78 Amelard R, Glaister J, Wong A, Clausi DA (2015) High-Level Intuitive Features (HLIFs) for intuitive skin lesion description. IEEE Trans Biomed Eng 62:820–831. https://doi.org/10.1109/ TBME.2014.2365518 Argenziano G, Soyer HP, De Giorgi V, et al (2000) Interactive atlas of dermoscopy (Book and CD-ROM) Dermoscopy Atlas. https://www.dermoscopyatlas.com/. Accessed 22 Sep 2018 Atlas of Dermatology. https://www.dermnet.com/. Accessed 22 Sep 2018 Barata C, Ruela M, Francisco M et al (2014) Two systems for the detection of melanomas in dermoscopy images using texture and color features. IEEE Syst J 8:965–979. https://doi.org/10. 1109/JSYST.2013.2271540 Barata C, Celebi ME, Marques JS (2015) Melanoma detection algorithm based on feature fusion. 2653–2656 Barzegari M, Ghaninezhad H, Mansoori P et al (2005) Computer-aided dermoscopy for diagnosis of melanoma. BMC Dermatol 5:1–4. https://doi.org/10.1186/1471-5945-5-8 Berseth M (2017) ISIC 2017: skin lesion analysis towards melanoma detection. 1–4 Borja-Cacho D, Matthews J (2008) Scratching the surface of skin development. NANO 6:2166– 2171. https://doi.org/10.1021/nl061786n.Core-Shell Cadenas JM, Garrido MC, Martínez R (2013) Feature subset selection filter–wrapper based on low quality data. 40:6241–6252. https://doi.org/10.1016/j.eswa.2013.05.051 Cavalcanti PG, Scharcanski J, Baranoski GVG (2013) Expert systems with applications a two-stage approach for discriminating melanocytic skin lesions using standard cameras. Expert Syst Appl 40:4054–4064. https://doi.org/10.1016/j.eswa.2013.01.002 Chandrashekar G, Sahin F (2014) A survey on feature selection methods. Comput Electr Eng 40:16–28. https://doi.org/10.1016/j.compeleceng.2013.11.024 Chang Y, Stanley RJ, Moss RH, Van Stoecker W (2005) A systematic heuristic approach for feature selection for melanoma discrimination using clinical images. Ski Res Technol 11:165–178. https:// doi.org/10.1111/j.1600-0846.2005.00116.x Chan JCW, Paelinckx D (2008) Evaluation of random forest and adaboost tree-based ensemble classification and spectral band selection for ecotope mapping using airborne hyperspectral imagery. Remote Sens Environ 112:2999–3011. https://doi.org/10.1016/j.rse.2008.02.011 Codella N (2015) Accurate and scalable system for automatic detection of malignant melanoma. https://doi.org/10.1201/b19107-11 Codella NCF, Gutman D, Celebi ME, et al (2018) Skin lesion analysis toward melanoma detection: a challenge at the international symposium on biomedical imaging (ISBI) 2016, hosted by the international skin imaging collaboration (ISIC). In: Proceedings of the international symposium on biomedical imaging 2018, pp 168–172. https://doi.org/10.1109/ISBI.2018.8363547 Criminisi A (2011) Decision forests: a unified framework for classification, regression, density estimation, manifold learning and semi-supervised learning. Found Trends® Comput Graph Vis 7:81–227. https://doi.org/10.1561/0600000035
318
L. Singh et al.
D’Orazio J, Jarrett S, Amaro-Ortiz A, Scott T (2013) UV radiation and the skin. Int J Mol Sci 14:12222–12248. https://doi.org/10.3390/ijms140612222 Dalila F, Zohra A, Reda K, Hocine C (2017) Segmentation and classification of melanoma and benign skin lesions. Opt—Int J Light Electron Opt 140:749–761. https://doi.org/10.1016/j.ijleo. 2017.04.084 DermIS. https://www.dermis.net/dermisroot/en/home/index.htm. Accessed 22 Sep 2018 Di Leo G, Fabbrocini G, Paolillo A, et al (2009) Towards an automatic diagnosis system for skin lesions: estimation of blue-whitish veil and regression structures. In: 2009 6th international multiconference on systems signals devices, SSD 2009, vol 1, pp 1–6. https://doi.org/10.1109/SSD. 2009.4956770 Ercal F, Chawla A, Stoecker WV, et al (1994) Neural network diagnosis of malignant melanoma from color images. IEEE Trans Biomed Eng 41:837–845. https://doi.org/10.1109/10.312091 Erickson BJ, Korfiatis P, Akkus Z, Kline TL (2017) Machine learning for medical imaging. RadioGraphics 37:505–515. https://doi.org/10.1148/rg.2017160130 Garnavi R, Aldeen M, Bailey J (2012) Computer-aided diagnosis of melanoma using border- and wavelet-texture analysis. IEEE Trans Inf Technol Biomed 16:1239–1252. https://doi.org/10.1109/ TITB.2012.2212282 Gehler PV, Schölkopf B (2009) An introduction to kernel learning algorithms. Max-PlanckGesellschaft, Biol Kybern 25–48. https://doi.org/10.1109/72.914517 Giotis I, Molders N, Land S, et al (2015) MED-NODE: a computer-assisted melanoma diagnosis system using non-dermoscopic images. Expert Syst Appl 42:6578–6585. https://doi.org/10.1016/ j.eswa.2015.04.034 Haralick RM, Shanmugam K, Dinstein I (1973) Textural features for image classification. IEEE Trans Syst Man Cybern SMC 3:610–621 Huberty CJ (2018) Discriminant analysis. 45:543–598 Dermofit Image Library—Edinburgh Innovations. https://licensing.eri.ed.ac.uk/i/software/der mofit-image-library.html. Accessed 22 Sep 2018 Jaglan P, Dass R, Duhan M (2019) A comparative analysis of various image segmentation techniques. In: Proceedings of 2nd International Conference on Communication, Computing and Networking. Springer, Singapore, pp 359–374 Jaleel JA, Salim S (2013) Computer aided detection of skin cancer. 1137–1142 Jerant AF, Johnson JT, Demastes Sheridan C, Caffrey TJ (2000) Early detection and treatment of skin cancer. Am Fam Phys 62 Kasmi R, Mokrani K (2016) Classification of malignant melanoma and benign skin lesions: implementation of automatic ABCD rule. 10:448–455. https://doi.org/10.1049/iet-ipr.2015.0385 Kaufman HL (2005) The melanoma book: a complete guide to prevention and treatment. Gotham Korotkov K, Garcia R (2012) Computerized analysis of pigmented skin lesions: a review. Artif Intell Med 56:69–90. https://doi.org/10.1016/j.artmed.2012.08.002 Lee CH, Gutierrez F, Dou D (2011) Calculating feature weights in naive Bayes with KullbackLeibler measure. In: Proceedings of the IEEE international conference on data mining, ICDM, pp 1146–1151. https://doi.org/10.1109/ICDM.2011.29 Lew MS (2013) Principles of visual information retrieval. p 152 Lowe NJ (2006) An overview of ultraviolet radiation, sunscreens, and photo-induced dermatoses. Dermatol Clin 24:9–17. https://doi.org/10.1016/j.det.2005.08.001 Mark Elwood J, Jopson J (1997) Melanoma and sun exposure: an overview of published studies. Int J Cancer 73:198–203. https://doi.org/10.1002/(SICI)1097-0215(19971009)73:2%3c198::AIDIJC6%3e3.0.CO;2-R Ma L, Staunton RC (2013) Analysis of the contour structural irregularity of skin lesions using wavelet decomposition. Pattern Recognit 46:98–106. https://doi.org/10.1016/j.patcog. 2012.07.001 Mehta P, Shah B (2016) Review on techniques and steps of computer aided skin cancer diagnosis. Procedia Comput Sci 85:309–316. https://doi.org/10.1016/j.procs.2016.05.238
Automated CAD System for Skin Lesion Diagnosis: A Review
319
Melanoma M, The FOR, The ROF (2007) Malignant melanoma in the 21st century, part 1: epidemiology, risk factors, screening, prevention, and diagnosis. 82:364–380 Mendonca T, Ferreira PM, Marques JS, et al (2013) PH2 —a dermoscopic image database for research and benchmarking. In: 2013 annual international conference of the IEEE engineering in medicine and biology society, pp 5437–5440. https://doi.org/10.1109/EMBC.2013.6610779 Møllersen K, Hindberg K, Schopf TR et al, Skin Lesion 9 improved diagnostics for general practice by computer-aided diagnostics Okur E, Turkan M (2018) A survey on automated melanoma detection. Eng Appl Artif Intell 73:50–67. https://doi.org/10.1016/j.engappai.2018.04.028 Oliveira RB, Filho ME, Ma Z et al (2016a) Computational methods for the image segmentation of pigmented skin lesions: a review. Comput Methods Programs Biomed 131:127–141. https://doi. org/10.1016/j.cmpb.2016.03.032 Oliveira RB, Marranghello N, Pereira AS, Tavares JMR (2016b) A computational approach for detecting pigmented skin lesions in macroscopic images. 1–30 Pathan S, Prabhu KG, Siddalingaswamy PC (2018) Techniques and algorithms for computer aided diagnosis of pigmented skin lesions—a review. Biomed Signal Process Control 39:237–262. https://doi.org/10.1016/j.bspc.2017.07.010 Pereira AS, Manuel J, Tavares RS (2016) Computational methods for the image segmentation of pigmented skin lesions: a review. 131:127–141. https://doi.org/10.1016/j.cmpb.2016.03.032 Rastgoo M, Garcia R, Morel O, Marzani F (2015) Automatic differentiation of melanoma from dysplastic nevi. Comput Med Imaging Graph 43:44–52. https://doi.org/10.1016/j.compmedimag. 2015.02.011 Rodríguez R, Sossa JH (2017) Mathematical techniques for biomedical image segmentation. Elsevier Rourrur ASG, Rvise SNS (1982) Alternative k-nearest neighbour rules in supervised pattern recognition. 136:15–27 Schaefer G, Krawczyk B, Celebi ME, Iyatomi H (2014) An ensemble classification approach for melanoma diagnosis. Memetic Comput 6:233–240. https://doi.org/10.1007/s12293-014-0144-8 Search results. https://www.dermquest.com/results/?q=Malignantmelanoma. Accessed 22 Sep 2018 Shimizu K, Iyatomi H, Celebi ME et al (2015) Four-class classification of skin lesions with task decomposition strategy. 62:274–283 Siegel R, Miller KD, Ahmedin J (2017) Cancer statistics . Ca Cancer J 67:7–30. https://doi.org/10. 3322/caac.21387 Singh D, Gautam D, Ahmed M (2014) Detection techniques for melanoma diagnosis: a performance evaluation. In: 2014 international conference on signal propagation and computer technology ICSPCT 2014, pp 567–572. https://doi.org/10.1109/ICSPCT.2014.6884948 Singh BK, Verma K, Panigrahi L, Thoke AS (2017) Integrating radiologist feedback with computer aided diagnostic systems for breast cancer risk prediction in ultrasonic images: An experimental investigation in machine learning paradigm. Expert Syst Appl 90:209–223. https://doi.org/10. 1016/j.eswa.2017.08.020 Slominski A, Wortsman J (2013) Neuroendocrinology of the Skin1. Endocr Rev 21:457–487. https:// doi.org/10.1210/edrv.21.5.0410 Slominski AT, Zmijewski MA, Skobowiat C, et al (2012) Sensing the environment: regulation of local and global homeostasis by the skin’s neuroendocrine system Society AC (2011) Colorectal cancer facts & figures 2011–2013 Tharwat A (2018) Classification assessment methods. Appl Comput Informatics. https://doi.org/ 10.1016/j.aci.2018.08.003 Vanaja S (2014) Analysis of feature selection algorithms on classification: a survey. 96:28–35 (2017) What is melanoma skin cancer? Am Cancer Soc 1–98 Yu L, Member S, Chen H, et al (2016) Automated melanoma recognition in dermoscopy images via very deep residual networks. 0062:1–11. https://doi.org/10.1109/TMI.2016.2642839
320
L. Singh et al.
Zitová B, Flusser J (2003) Image registration methods: a survey. Image Vis Comput 21:977–1000. https://doi.org/10.1016/S0262-8856(03)00137-9 Zortea M, Schopf TR, Thon K et al (2014) Artificial intelligence in medicine performance of a dermoscopy-based computer vision system for the diagnosis of pigmented skin lesions compared with visual evaluation by experienced dermatologists. Artif Intell Med 60:13–26. https://doi.org/ 10.1016/j.artmed.2013.11.006
Medical Diagnosis of Coronary Artery Disease Using Fuzzy Rule-Based Classification Approach Namrata Singh and Pradeep Singh
Abstract Coronary Artery Disease (CAD) is one of the leading causes of morbidity and mortality worldwide including India. Although recent advances in modern medical science have led to better diagnosis and treatment of CAD, yet its early detection is still a challenge. Fuzzy classification approaches are used to deal with uncertainty inherent in medical field. These fuzzy rule-based systems are extremely effective tools in disease diagnosis as they are capable to develop potential linguistic models. The aim of this paper is to initially develop a fuzzy rule-based classification system (FRBCS) based on clinical and epidemiological variables of patients and then to determine its accuracy in the diagnosis of CAD. The membership functions for medical attributes were chosen after extensive review of related literature. The rules were formulated as per the opinion of expert physicians. The present work describes the risk factors accountable for CAD, fuzzy modeling of clinical variables, rule evaluation and defuzzification of the fuzzified outputs to crisp values. The accuracy of the proposed fuzzy if–then rule classification system is 89%. Further, the present approach can assist medical practitioners in diagnosing CAD more precisely based on the fuzzy rules. Keywords Fuzzy rule-based classification systems · Medical decision support · Coronary artery disease diagnosis · Classification · Cardiovascular diseases
1 Introduction The development of Coronary artery disease (CAD) is described as the slow growth of atherosclerotic plaques in coronary arteries, causing luminal stenosis. This further leads to occlusion, thus resulting in myocardial infarction (MI) (Tsipouras et al. N. Singh (B) · P. Singh Department of Computer Science and Engineering, National Institute of Technology, Raipur 492001, Chhattisgarh, India e-mail: [email protected] P. Singh e-mail: [email protected] © Springer Nature Singapore Pte Ltd. 2021 A. A. Rizvanov et al. (eds.), Advances in Biomedical Engineering and Technology, Lecture Notes in Bioengineering, https://doi.org/10.1007/978-981-15-6329-4_27
321
322
N. Singh and P. Singh
2008). Probably the most common cause of sudden cardiac death is attributed to cardiovascular diseases worldwide. Considering the epidemiology and pathophysiology of CAD, the early identification and effective modification of risk factors of coronary heart disease, the prevention of its growth and further complications, its diagnosis and timely treatment are of utmost importance to combat the disease (Alizadehsani et al. 2018). One of the widely used gold standard technique for the diagnosis of CAD is coronary angiography (CA). Since CA is an invasive and expensive method, therefore for the diagnosis of CAD other noninvasive techniques are being utilized in the clinical practice. In the past years, numerous computer-aided diagnosis methodologies related to CAD have been proposed in the research works (Alizadehsani et al. 2016; Anooj 2012). Thus, noninvasive methods capable of predicting the presence of CAD using easily acquired attributes and providing an understanding of the decisions made would be of immense clinical significance (Verma et al. 2016, 2018). Fuzzy modeling can be used to handle the fuzziness inherent in biomedical problems and its integration with data mining gives the desired interpretation for the acquired decisions. Many of the state-of-the-art fuzzy rule-based classification approaches and computational methods have been developed for detecting CAD. Sabahi (2018) proposed a novel bimodal fuzzy analytic hierarchy process (BFAHP) for dealing with uncertainty in multiple criteria decision-making (MCDM). This approach calculates the fuzzy validity by amalgamating the validities of suitable risk factors based on collective intelligence and expert knowledge. The Bayesian formulation is used to compute the fuzzy probability of risk factors. BFAHP is applied to a real dataset for CHD risk assessment which identifies diastolic blood pressure and high-density lipoprotein as the significant risk factors. Sanz et al. (2014) developed a system which predicts the risk of cardiovascular disease within the next 10 years. The approach provides both an interpretable model and a diagnosis for describing the decision, helpful for the physicians. The proposed methodology aggregates FRBCS with interval-valued fuzzy sets that consist of modeling of linguistic labels of the learner and applying genetic tuning for optimization of FRBCS. The recent Clinical Decision Support System (CDSS) proposed by Nazari et al. (2018) utilizes Fuzzy Inference System (FIS) and Fuzzy Analytic Hierarchy Process (FAHP) for assessment of heart disease. Pal et al. (2012) developed a system for screening and early detection of CAD by utilizing easily accessible clinical parameters and laboratory tests. Muthukaruppan and Er (2012) proposed a hybrid particle swarm optimization-(PSO) based fuzzy expert system for the CAD diagnosis. Singh and Singh (2019) utilized various machine learning techniques for cardiac arrhythmia classification. Further, rule-based classification proves to be immensely helpful for classification of other diseases too (Singh and Singh 2017). In this contribution, the proposed FRBS was constructed with formulation of 30 fuzzy rules and Mamdani fuzzy inference approach was applied as a decision-making technique for CAD classification.
Medical Diagnosis of Coronary Artery Disease …
323
In this paper, we propose a fuzzy rule-based classification system (FRBCS) for the diagnosis of CAD. Section 2 describes the fuzzy rule-based methodology, Sect. 3 illustrates about results and discussion, and finally the conclusion is provided in Sect. 4.
2 Fuzzy Rule-Based Methodology The methodology proposed for the creation of fuzzy decision support system utilizes the Mamdani approach. In order to generate a diagnosis, an unidentified case is given to the obtained decision support system. The fuzzy expert system is described in the following steps.
2.1 Variables Selection The variables selected as input in the fuzzy expert system were 6 namely, age, fasting blood sugar (FBS), triglyceride level (TGL), low-density lipoprotein (LDL), systolic blood pressure (SBP), and high-density lipoprotein (HDL). Further, these input parameters were fuzzified ranging from 0 to 1.
2.2 Fuzzification The trapezoidal membership function is used for the given interval range because of its capacity and generality to incorporate more fuzzy information.
2.3 Knowledge Base (IF–THEN Rules Formulation) A set of rules was used for relating the fuzzy variables to the outcome classification. For modeling of linguistic variables, the Mamdani type fuzzy rule base is utilized, formulated in the following way: IF … Clinical variable 1 AND Clinical variable 2… THEN… Normal or CAD. Some of clinical fuzzy rules are illustrated below: (a) If (Age is Young) and (FBS is Good) and (TGL is Normal) and (LDL is Normal) and (SBP is Normal) and (HDL is High) then (Output is No) (b) If (Age is Young) and (FBS is Bad) and (TGL is Moderate) and (LDL is Moderate) and (SBP is Middle) and (HDL is Moderate) then (Output is Yes)
324
N. Singh and P. Singh
3 Results and Discussion 3.1 Patients The coronary artery disease data (Z-Alizadeh Sani Dataset) used in this paper is obtained from the UCI machine learning (Dheeru and Karra Taniskidou 2017) database containing 303 patient samples, where each instance had 54 features. The features were arranged in four groups, namely symptom and examination, demographic, laboratory and echo, and ECG features. Each sample was classified into two classes, viz. class 1 (normal) and class 2 (CAD) having 87 and 216 subjects, respectively. If a patient’s coronary artery diameter narrowing is greater than or equal to 50%, then he/she is categorized as having CAD, or otherwise as normal. Out of 54 features, 6 suitable risk factors accountable for CAD are recognized and used as input to the FIS. These factors are age, fasting blood sugar (FBS), triglyceride level (TGL), low-density lipoprotein (LDL), systolic blood pressure (SBP), and highdensity lipoprotein (HDL). We used membership functions for medical variables according to literature review (Ross 2010; Mohammadpour et al. 2015; D’Acierno et al. 2013; Peña-Reyes and Sipper 2002). For each of these 6 variables, we took 3 fuzzy sets namely for age (young, middle, and old), FBS (good, acceptable, and bad), TGL (normal, moderate, and high), LDL (normal, middle, and high), SBP (low, middle, and high), and LDL (low, middle, and high). The trapezoidal membership function is applied for each fuzzy set as shown in Fig. 1, and the ranges of linguistic variables are given in Table 1. In Table 1, SI denotes the support interval and CI denotes the core interval of the range of different levels of linguistic variables for various risk factors. Further, Fig. 2 demonstrates the output value for the given input values.
3.2 Comparison with the Prior Work Comparative performance analysis of the proposed fuzzy rule-based classification system in classifying CAD to existing works is shown in Table 2. In Fig. 3, the confusion matrix demonstrates the tabular fuzzy rule-based system classification results summary of the true or actual labels versus the predicted ones. The classifier evaluation has been performed on test data using three evaluation metrics viz. accuracy, sensitivity, and specificity. The accuracy, sensitivity, and specificity of the classification system are 89%, 98.04%, and 86.49%, respectively.
Medical Diagnosis of Coronary Artery Disease … 1
0.8 0.6 0.4 0.2 0 20
1
40
30
Normal
50
0.6 0.4 0.2 0 100
200
(i)
(ii) High
1
0.6 0.4 0.2 0 250
300
Moderate
0.6 0.4 0.2 0 120
140
(iv)
Middle
High
0.8 0.6 0.4 0.2 0 120
130
140
150
Degree of membership
Degree of membership
(iii) Normal
160
LDL
TGL
1
High
0.8
100
400
350
Normal
400
300
FBS
Moderate
200
0.8
Age
0.8
150
Bad
Good Acceptable
70
60
Degree of membership
Degree of membership
Old
Middle
Young
Degree of membership
Degree of membership
1
325
1
Normal
Middle
High
0.8 0.6 0.4 0.2 0 30
40
50
60
SBP
HDL
(v)
(vi)
70
80
Fig. 1 Graph of membership function a age, b fasting blood sugar, c triglyceride level, d LDL, e systolic blood pressure, f HDL
4 Conclusion We have presented a fuzzy rule-based classification system for CAD diagnosis. One of the most significant results of this work is the classification of CAD while balancing the tradeoff between interpretability and accuracy. The major findings of
70–110
=60
326 N. Singh and P. Singh
Medical Diagnosis of Coronary Artery Disease …
327
Fig. 2 Delivered output based on different inputs
this approach are the selection of suitable shape and number of membership functions and fuzzy rules which improve the classification accuracy of the overall system, thus leading to higher quality of precision. The classification accuracy depends upon various factors such as number of input features, type of membership functions, appropriate formulation and combination of fuzzy rules. These rules describe the relationship between the symptoms and the diagnosis outcome in a more accurate and understandable way. Thus, the proposed fuzzy expert system achieves an accuracy of 89%, sensitivity 98.04%, and specificity 86.49%. Further, rule pruning and rule minimization that affect the system’s decision-making capabilities can be considered as future work.
328
N. Singh and P. Singh
Table 2 Comparison of performances of FRBCS in CAD diagnosis Author (year)
Approach
Dataset
Performance Parameters
Sabahi (2018)
BFAHP
Real
ACC
85.91
Cleveland
87.31
Hungarian
86.57
Switzerland
85.22
Long Beach VA Sanz et al. (2014)
IVFS_Amp+ K α (Chi)
85.62
Real
IVFS_Amp+ K α (FH-GBML) Pal et al. (2012)
Muthukaruppan and Er (2012)
Paul et al. (2018)
Fuzzy expert system
PSO-based fuzzy expert system
Adaptive weighted FRBS
Real
Hungarian and Cleveland
Cleveland
ACC
73.82
ACC
73.71
ACC
84.20
SPEC
83.33
SENS
95.85
ACC
93.27
SPEC
93.3
SENS
93.2
ACC
92.31
Hungarian
95.56
Switzerland
89.47
Long Beach VA
91.80
Heart Reddy and Khare (2017)
OFBAT + RBFL
92.68
Cleveland
ACC
68.55
Hungarian
66.67
Switzerland
78
Priyatharshini and Chitrakala (2019)
SL-FRBS
Cleveland
ACC SENS
91.5
Proposed work
FRBCS
Z-Alizadeh Sani
ACC
89
SPEC
86.49
SENS
98.04
Fig. 3 Confusion matrix of FRBCS
90.7
True Label
Predicted Labels Normal
CAD
Normal
50
1
CAD
20
128
Medical Diagnosis of Coronary Artery Disease …
329
References Alizadehsani R, Zangooei MH, Hosseini MJ, Habibi J, Khosravi A, Roshanzamir M, Khozeimeh F, Sarrafzadegan N, Nahavandi S (2016) Coronary artery disease detection using computational intelligence methods. Knowl-Based Syst 109:187–197 Alizadehsani R, Hosseini MJ, Khosravi A, Khozeimeh F, Roshanzamir M, Sarrafzadegan N, Nahavandi S (2018) Non-invasive detection of coronary artery disease in high-risk patients based on the stenosis prediction of separate coronary arteries. Comput Methods Programs Biomed 162:119–127 Anooj PK (2012) Clinical decision support system: risk level prediction of heart disease using weighted fuzzy rules. J King Saud Univ—Comput Inf Sci 24:27–40 D’Acierno A, Esposito M, De Pietro G (2013) An extensible six-step methodology to automatically generate fuzzy DSSs for diagnostic applications. BMC Bioinformatics 14:S4 Dheeru D, Karra Taniskidou E (2017) UCI machine learning repository. https://archive.ics.uci. edu/ml Mohammadpour RA, Abedi SM, Bagheri S, Ghaemian A (2015) Fuzzy rule-based classification system for assessing coronary artery disease. Comput Math Methods Med 2015:1–8 Muthukaruppan S, Er MJ (2012) A hybrid particle swarm optimization based fuzzy expert system for the diagnosis of coronary artery disease. Expert Syst Appl 39:11657–11665 Nazari S, Fallah M, Kazemipoor H, Salehipour A (2018) A fuzzy inference- fuzzy analytic hierarchy process-based clinical decision support system for diagnosis of heart diseases. Expert Syst Appl 95:261–271 Pal D, Mandana KM, Pal S, Sarkar D, Chakraborty C (2012) Fuzzy expert system approach for coronary artery disease screening using clinical parameters. Knowl-Based Syst 36:162–174 Paul AK, Shill PC, Rabin MRI, Murase K (2018) Adaptive weighted fuzzy rule-based system for the risk level assessment of heart disease. Appl Intell 48:1739–1756. https://doi.org/10.1007/s10 489-017-1037-6 Peña-Reyes CA, Sipper M (2002) Combining evolutionary and fuzzy techniques in medical diagnosis. In: Schmitt M, Teodorescu H, Jain A, Jain A, Jain S, Jain LC (eds) Computational intelligence processing in medical diagnosis. Studies in fuzziness and soft computing. Physica, Heidelberg, pp 391–426 Priyatharshini R, Chitrakala S (2019) A self-learning fuzzy rule-based system for risk-level assessment of coronary heart disease. IETE J Res 65:288–297. https://doi.org/10.1080/03772063.2018. 1431062 Reddy GT, Khare N (2017) An efficient system for heart disease prediction using hybrid OFBAT with rule-based fuzzy logic model. J Circuits, Syst Comput 26:1750061. https://doi.org/10.1142/ S021812661750061X Ross TJ (2010) Fuzzy logic with engineering applications. Wiley, Chichester, UK Sabahi F (2018) Bimodal fuzzy analytic hierarchy process (BFAHP) for coronary heart disease risk assessment. J Biomed Inform 83:204–216 Sanz JA, Galar M, Jurio A, Brugos A, Pagola M, Bustince H (2014) Medical diagnosis of cardiovascular diseases using an interval-valued fuzzy rule-based classification system. Appl Soft Comput 20:103–111 Singh N, Singh P (2017) Rule based approach for prediction of chronic kidney disease: a comparative study. Biomed Pharmacol J 10:867–874 Singh N, Singh P (2019) Cardiac arrhythmia classification using machine learning techniques. In: Engineering vibration, communication and information processing. Springer, Singapore. pp 469–480 Tsipouras MG, Exarchos TP, Fotiadis DI, Kotsia AP, Vakalis KV, Naka KK, Michalis LK (2008) Automated diagnosis of coronary artery disease based on data mining and fuzzy modeling. IEEE Trans Inf Technol Biomed 12:447–458
330
N. Singh and P. Singh
Verma L, Srivastava S, Negi PC (2016) A hybrid data mining model to predict coronary artery disease cases using non-invasive clinical data. J Med Syst 40:178 Verma L, Srivastava S, Negi PC (2018) An intelligent noninvasive model for coronary artery disease detection. Complex Intell Syst 4:11–18
Segmentation of Lungs in Thoracic CTs Using K-means Clustering and Morphological Operations Satya Prakash Sahu, Rahul Kumar, Narendra D. Londhe, and Shrish Verma
Abstract Lung segmentation is the most essential and significant step for computeraided diagnosis (CAD) systems for the detection of cancer in the lung region at the prior phase. The efficacy of CAD system is highly dependent on how the lungs are properly and accurately segmented. Effective lung segmentation reduces the various challenges for the CAD system to detect the juxtapleural nodules. This paper emphasizes the segmentation of lungs in CT images using the K-means approach with automatic threshold and morphological operations. Total ten subjects have been taken from the public dataset LIDC-IDRI including two subjects of juxtapleural nodules. The proposed method achieved a performance of 97.522% overlap ratio and 0.9685 Jaccard’s similarity index values. Keywords CAD system · Lung segmentation · Computed tomography · K-means · Nodule detection · Morphological operations
1 Introduction Among the different types of cancer, lung cancer remains the most common cancer worldwide. There were 1.8 million new lung cancer cases estimated to occur in 2012 (Ferlay et al. 2013). In the US, from the American Cancer Society, Cancer Facts and Figures 2017 (American Cancer Society 2017), there were approximately 222,500 new cases of cancer and 155,870 related deaths that accounts for one out of four cancer-related deaths. Lung cancer is leading to the highest mortality rate; this has urged researchers to perform a diagnosis at the early stages. The survival rate may be greatly enhanced, i.e., by 70–80% if lung cancer is detected in the first stage (Swensen et al. 2002). The various challenges in lung segmentation are due to inhomogeneities in the lung region and similar densities in other pulmonary structures like veins, arteries, bronchi, and bronchioles. However, the most challenging task for researchers is to detect S. P. Sahu (B) · R. Kumar · N. D. Londhe · S. Verma National Institute of Technology, Raipur, India e-mail: [email protected] © Springer Nature Singapore Pte Ltd. 2021 A. A. Rizvanov et al. (eds.), Advances in Biomedical Engineering and Technology, Lecture Notes in Bioengineering, https://doi.org/10.1007/978-981-15-6329-4_28
331
332
S. P. Sahu et al.
the nodules attached to pleural surface (i.e., juxtapleural). If the lung segmentation process does not define the lung boundaries correctly, the juxtapleural nodule may be missed. Armato et al. showed that approximately 5–17% true lung nodules are not detected due to poor and inaccurate lung segmentation (Armato and Sensakovic 2004). Thus, there is a need for effective lung segmentation technique for the enhancement in the efficiency of CAD system, i.e., truly capable of identifying the true lung nodules. In this paper, a hybrid method using k-means clustering and automatic thresholding has been proposed for lung segmentation to delineate the boundary of lungs and to extract the region of interest (ROI) area. Then, the broken borders are corrected by applying morphological closing operations to include the juxtaplueral nodules to minimize the chance of over-segmentation. Figure 1 shows the process flow diagram for the adopted method.
2 Related Work Threshold-based approach is the most popular method for image segmentation, which utilizes the intensity (gray level) value of the image. Since the lung regions are having lower gray level (−500 HU approximately) with respect to other anatomical structures in the thorax region, hence, with the application of optimum threshold, the segmentation of lungs has been adopted by a number of authors in their research articles. Hu et al. (2001) used iterative thresholding and some morphological operations for lung segmentation. A similar threshold-based approach was given by Gao et al. (2007), i.e., region growing and morphological smoothing. Wei et al. (2009) proposed a threshold using histogram analysis to segment the lung region. Adaptive fuzzy thresholding was given by Ye et al. (2009) in lung segmentation from CT data. Helen et al. (2011) enhanced the performance of 2D Otsu-based thresholding algorithm for the segmentation of pulmonary parenchyma in CT lung images using particle swarm optimization (PSO). Sahu et al. (2017, 2019) have given the hybrid method based on fuzzy clustering, thresholding, and morphological operations. Active contour-based method (ACM) is the next significant method that has been greatly utilized and used in various research studies (Xu and Prince 1998; Wang et al. 2010; Chan and Vese 2001; Cui et al. 2013; Athertya and Kumar 2014). This method relies on the minimization of energy function through dynamic contour iteratively. However, the problem of initialization and convergence are the challenges with this method which had been addressed by a number of authors in their articles. Xu and Prince (1998) worked for better convergence and introduced an external force, i.e., gradient vector flow (GVF) to guide the contour. Wang et al. (2010) has given the improved version for these issues through normally biased GVF. Chan and Vese (2001) has given ACM without edges and used curve evolution techniques for the initialization problem. Cui et al. (2013) given the ACM based on statistics of a local region for the segmentation of medical images. The fully automatic method
Segmentation of Lungs in Thoracic CTs …
333
using ACM is exploited by Athertya and Kumar (2014) that reduces the interaction of the user. The other useful technique that uses the information of earlier shape and parameters of lungs for obtaining the ROI from the thorax region is shape-based segmentation. The shape parameters such as contour or edges and points of previous data are characterized for the derivation of variational energy framework for the segmentation of lungs. Annangi et al. (2010) proposed a region-based active contour method with the use of prior shape and low-level features to handle the issue of local minima. Kockelkorn et al. (2010) proposed interactive techniques for lung segmentation in CT data with severe abnormalities where prior shape term are trained using k-NN classifier thereby correction in classification results. Sofka et al. (2011) adopted a pattern recognition technique with the combination of statistical shape model and anatomical information for the robust lung segmentation. A novel approach is given by Sun et al. (2011) based on the robust active shape model, and further, a constrained optimal surface finding method has been used to adapt the initial segmentation result. For the initialization of active shape models, Gill et al. (2014) gave the feature-based atlas approach for segmentation in CT images.
3 Material and Methods 3.1 Data The public dataset of the Lung Imaging Database consortium-Image Database Resource Initiative (LIDC-IDRI) has been accessed for the experimental database (Armato et al. 2015). This dataset contains 1018 cases of Lung CT (Computer Tomography) scans (Armato et al. 2011). The LIDC dataset also contains the annotated information in the attached XML file provided by four radiologists (experts) for each CT scan (Clark et al. 2013). For the proposed method, ten patients have been acquired through the LIDC dataset with the following specifications: DICOM Image of size 512 × 512, the image intensity value from −600 to 1600 Hounsfield Unit (HU), tube current of 265–570 mA and tube voltage of 120 kVp.
3.2 Data Preprocessing The experimental database contains CT scans of ten patients with some cases of CTs having juxtapleural lung nodules. The contoured outline or boundary tracing for lung region for each patient of all the slices in different cases is done by an expert as the representation of ground truth. For improving the accuracy in the further operations, the pixel values of the image are converted to double format followed by image enhancement for better contrast. Pixel value of the image needs to be represented in
334
S. P. Sahu et al.
HU value for setting appropriate window level and window width are set through the. As per the DICOM file, the Rescale Slope and Rescale Intercept are done by a linear transformation. Linear transformation leads to an effective range of HU values for the images; afterward, the HU value greater than the upper range will be shown in white, and HU value that is lesser than the lower range will be shown in black. Then the operations of image processing for the conversion of 16-bits image into an 8-bits image will be applied.
3.3 Segmentation of Lungs The process of lung segmentation is proposed as follows: A. K-means clustering is applied to the input CT image thereby generating the grayscale-masked image. (Algorithm and derivations explained in Sect. 3.3.1) B. The automatic thresholding method applied to the output of K-means masks for the conversion into binary, and then the background subtraction for ROI is performed. (explained in Sect. 3.3.2) C. The large airways and other vessels within the lung region are removed by the hole-filling algorithm. Let S be the set whose boundary pixel is labeled by 1 s (i.e., 8-connected boundaries) and each boundary enclosing a hole (background region). With the given point p in each hole, the objective is to fill the holes by 1 s. This algorithm fills the holes in an image through the process of dilation, intersection, and complementation. The algorithm is followed by morphological reconstruction applied to the lung portion and the lighter border is also removed by reducing the overall intensity of the border structures. D. For handling the juxtapleural cases, the morphological closing operations are applied in all lung CT images. E. The final segmented lung regions are obtained by masking the output image with an input grayscale image.
3.3.1
K-means Algorithm
K-means is an unsupervised and simple clustering approach given by Hartigan and Wong (1979). The procedure follows an easy way of partitioning a given dataset (nobject or data points) into a certain number of k-clusters. This approach is utilized here for the segmentation process since it classifies the objects into some desired number of groups or clusters based on some selected features. The process of clustering is minimizing the sum of squares of distance iteratively from the considered data points to the corresponding centroid of the cluster. Given a set of data points (x 1 , x 2 ,…, x n ), where each data point is having ddimension real vector, k-means clustering objective is to partition the n-data points
Segmentation of Lungs in Thoracic CTs …
LIDC-IDRI Dataset
Selection of Suitable CT Scans
Preprocessing: image enhancement, linear transformation, setting window level and width, etc. Apply K-means clustering to generate gray-scale mask
Segmented Lungs
Output image masked with input image
335
Morphological closing for handling Juxtapleural cases
Removal of large airways&vessel s by Hole filling algorithm
Automatic thresholds for binary image
Fig. 1 Process flow diagram
into k(≤n) clusters or groups G = {G1 , G2 , …., Gk } to minimize the sum of squares of distance (squared Euclidean distance) within a cluster. The data points are clustered by minimizing the following objective function (F) around the centroids (ci , for i = 1, 2, …, k) F=
k
(x − ci )2
(1)
i=1 x∈G i
where ci is the mean point or centroid of all points x ∈ G i . The algorithm is as follows: 1. Intensity distribution of the image is computed for various intensities. a. for each data point x i : i. nearest centroid cz is calculated, i.e., Euclidean distance between data point x i (intensity of an instance point) and all the cluster centers cj (centroid intensities) cz = arg min ||xi − c j ||2
(2)
ii. data point x i is assigned to cluster j, whose Euclidean distance is minimum from the jth cluster center among all the cluster centers. b. for each cluster j = 1,2 ,…, k 2. Initialize cluster centers (centroids) c1 , c2 , …, ck with k random intensities. 3. Repeat until convergence or shifting of cluster centers are observed
336
S. P. Sahu et al.
i. new centroid intensity cj , computed as the mean of all points x i assigned to cluster j in above step using cj =
nj 1 xi n j i=1
(3)
where nj represents the number of data points in jth cluster. 4. Stop, when no data points were reassigned, otherwise repeat step 3 above. 3.3.2
Thresholding
Otsu’s method is utilized here for automatic thresholding (Otsu 1975). This method finds the gray-level value, t* (estimated threshold), which minimizes the weighted intra-class variance and maximizes the inter-class variance. The algorithm assumes that the histogram of images is bimodal. Let the pixels of the image be represented in as 1, 2, …, L gray levels. The normalized histogram can be viewed as a probability distribution (pj ): L p j = n j N , p j ≥ 0, pj = 1
(4)
j=1
where nj denotes the number of a pixel having the gray levelj and the total number of pixel N = n1 + n2 + … + nL . Assuming the threshold at ‘t’, the pixels of an image will be classified in class C o (object as white) and C b (background as black) then, C b = [0,1,2,…,t] and C o = [t + 1, t + 2,…,L − 1]. The class occurrence probabilities for background and object will be: qb =
t
p j = q(t)
(5)
j=1
qo =
L−1
p j = 1 − q(t)
(6)
j=t+1
and, gray-level mean value for background and object will be given by: μb =
t j=1
j p j qb = μ(t) q(t)
(7)
Segmentation of Lungs in Thoracic CTs …
μ0 =
L−1 j=t+1
337
μT − μ(t) j p j q0 = 1 − q(t)
(8)
where q(t) as per Eq. (5) is 0th order and t μ(t) = j p j is first-order cumulative moments histogram up to the tth level, j=1
and μT = μ(L) =
L
jpj.
j=1
The optimum threshold will be obtained by maximizing the inter-class variance σ B2 (t). The generalized optimal threshold t* is obtained by histogram analysis and testing each gray level for which the possibility of the threshold t that maximizes σ B2 (t) is given by: σ B2 (t∗) = max σ B2 1≤t